Google Git Google Git
Sign in
llvm/llvm-project/refs/heads/users/void/array-bounds-accessed/./llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.384bit.ll
blob: eaea841d595ba083bbfc5c76e9855c4e0e70e70c [file] [edit]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgpu6.00 < %s | FileCheck -check-prefix=SI %s
; RUN: llc -mtriple=amdgpu8.02 < %s | FileCheck -check-prefix=VI %s
; RUN: llc -mtriple=amdgpu9.00 < %s | FileCheck -check-prefix=GFX9 %s
; RUN: llc -mtriple=amdgpu11.00 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s
; RUN: llc -mtriple=amdgpu11.00 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s
define <12 x float> @bitcast_v12i32_to_v12f32(<12 x i32> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v12f32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB0_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v11
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v9
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v7
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v5
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v3
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v1
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: .LBB0_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v12f32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB0_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v11, vcc, 3, v11
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_add_u32_e32 v9, vcc, 3, v9
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_add_u32_e32 v7, vcc, 3, v7
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_add_u32_e32 v5, vcc, 3, v5
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_add_u32_e32 v3, vcc, 3, v3
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_add_u32_e32 v1, vcc, 3, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: .LBB0_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v12f32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB0_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_u32_e32 v11, 3, v11
; GFX9-NEXT: v_add_u32_e32 v10, 3, v10
; GFX9-NEXT: v_add_u32_e32 v9, 3, v9
; GFX9-NEXT: v_add_u32_e32 v8, 3, v8
; GFX9-NEXT: v_add_u32_e32 v7, 3, v7
; GFX9-NEXT: v_add_u32_e32 v6, 3, v6
; GFX9-NEXT: v_add_u32_e32 v5, 3, v5
; GFX9-NEXT: v_add_u32_e32 v4, 3, v4
; GFX9-NEXT: v_add_u32_e32 v3, 3, v3
; GFX9-NEXT: v_add_u32_e32 v2, 3, v2
; GFX9-NEXT: v_add_u32_e32 v1, 3, v1
; GFX9-NEXT: v_add_u32_e32 v0, 3, v0
; GFX9-NEXT: .LBB0_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v12f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB0_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_nc_u32_e32 v11, 3, v11
; GFX11-NEXT: v_add_nc_u32_e32 v10, 3, v10
; GFX11-NEXT: v_add_nc_u32_e32 v9, 3, v9
; GFX11-NEXT: v_add_nc_u32_e32 v8, 3, v8
; GFX11-NEXT: v_add_nc_u32_e32 v7, 3, v7
; GFX11-NEXT: v_add_nc_u32_e32 v6, 3, v6
; GFX11-NEXT: v_add_nc_u32_e32 v5, 3, v5
; GFX11-NEXT: v_add_nc_u32_e32 v4, 3, v4
; GFX11-NEXT: v_add_nc_u32_e32 v3, 3, v3
; GFX11-NEXT: v_add_nc_u32_e32 v2, 3, v2
; GFX11-NEXT: v_add_nc_u32_e32 v1, 3, v1
; GFX11-NEXT: v_add_nc_u32_e32 v0, 3, v0
; GFX11-NEXT: .LBB0_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define inreg <12 x float> @bitcast_v12i32_to_v12f32_scalar(<12 x i32> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v12f32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc1 .LBB1_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: .LBB1_2: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v12f32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB1_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s27, s27, 3
; VI-NEXT: s_add_i32 s26, s26, 3
; VI-NEXT: s_add_i32 s25, s25, 3
; VI-NEXT: s_add_i32 s24, s24, 3
; VI-NEXT: s_add_i32 s23, s23, 3
; VI-NEXT: s_add_i32 s22, s22, 3
; VI-NEXT: s_add_i32 s21, s21, 3
; VI-NEXT: s_add_i32 s20, s20, 3
; VI-NEXT: s_add_i32 s19, s19, 3
; VI-NEXT: s_add_i32 s18, s18, 3
; VI-NEXT: s_add_i32 s17, s17, 3
; VI-NEXT: s_add_i32 s16, s16, 3
; VI-NEXT: .LBB1_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v12f32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB1_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_i32 s27, s27, 3
; GFX9-NEXT: s_add_i32 s26, s26, 3
; GFX9-NEXT: s_add_i32 s25, s25, 3
; GFX9-NEXT: s_add_i32 s24, s24, 3
; GFX9-NEXT: s_add_i32 s23, s23, 3
; GFX9-NEXT: s_add_i32 s22, s22, 3
; GFX9-NEXT: s_add_i32 s21, s21, 3
; GFX9-NEXT: s_add_i32 s20, s20, 3
; GFX9-NEXT: s_add_i32 s19, s19, 3
; GFX9-NEXT: s_add_i32 s18, s18, 3
; GFX9-NEXT: s_add_i32 s17, s17, 3
; GFX9-NEXT: s_add_i32 s16, s16, 3
; GFX9-NEXT: .LBB1_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v12f32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB1_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_i32 s23, s23, 3
; GFX11-NEXT: s_add_i32 s22, s22, 3
; GFX11-NEXT: s_add_i32 s21, s21, 3
; GFX11-NEXT: s_add_i32 s20, s20, 3
; GFX11-NEXT: s_add_i32 s19, s19, 3
; GFX11-NEXT: s_add_i32 s18, s18, 3
; GFX11-NEXT: s_add_i32 s17, s17, 3
; GFX11-NEXT: s_add_i32 s16, s16, 3
; GFX11-NEXT: s_add_i32 s3, s3, 3
; GFX11-NEXT: s_add_i32 s2, s2, 3
; GFX11-NEXT: s_add_i32 s1, s1, 3
; GFX11-NEXT: s_add_i32 s0, s0, 3
; GFX11-NEXT: .LBB1_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define <12 x i32> @bitcast_v12f32_to_v12i32(<12 x float> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v12i32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB2_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_f32_e32 v11, 1.0, v11
; SI-NEXT: v_add_f32_e32 v10, 1.0, v10
; SI-NEXT: v_add_f32_e32 v9, 1.0, v9
; SI-NEXT: v_add_f32_e32 v8, 1.0, v8
; SI-NEXT: v_add_f32_e32 v7, 1.0, v7
; SI-NEXT: v_add_f32_e32 v6, 1.0, v6
; SI-NEXT: v_add_f32_e32 v5, 1.0, v5
; SI-NEXT: v_add_f32_e32 v4, 1.0, v4
; SI-NEXT: v_add_f32_e32 v3, 1.0, v3
; SI-NEXT: v_add_f32_e32 v2, 1.0, v2
; SI-NEXT: v_add_f32_e32 v1, 1.0, v1
; SI-NEXT: v_add_f32_e32 v0, 1.0, v0
; SI-NEXT: .LBB2_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v12i32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB2_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f32_e32 v11, 1.0, v11
; VI-NEXT: v_add_f32_e32 v10, 1.0, v10
; VI-NEXT: v_add_f32_e32 v9, 1.0, v9
; VI-NEXT: v_add_f32_e32 v8, 1.0, v8
; VI-NEXT: v_add_f32_e32 v7, 1.0, v7
; VI-NEXT: v_add_f32_e32 v6, 1.0, v6
; VI-NEXT: v_add_f32_e32 v5, 1.0, v5
; VI-NEXT: v_add_f32_e32 v4, 1.0, v4
; VI-NEXT: v_add_f32_e32 v3, 1.0, v3
; VI-NEXT: v_add_f32_e32 v2, 1.0, v2
; VI-NEXT: v_add_f32_e32 v1, 1.0, v1
; VI-NEXT: v_add_f32_e32 v0, 1.0, v0
; VI-NEXT: .LBB2_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v12i32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB2_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f32_e32 v11, 1.0, v11
; GFX9-NEXT: v_add_f32_e32 v10, 1.0, v10
; GFX9-NEXT: v_add_f32_e32 v9, 1.0, v9
; GFX9-NEXT: v_add_f32_e32 v8, 1.0, v8
; GFX9-NEXT: v_add_f32_e32 v7, 1.0, v7
; GFX9-NEXT: v_add_f32_e32 v6, 1.0, v6
; GFX9-NEXT: v_add_f32_e32 v5, 1.0, v5
; GFX9-NEXT: v_add_f32_e32 v4, 1.0, v4
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v3
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v2
; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
; GFX9-NEXT: v_add_f32_e32 v0, 1.0, v0
; GFX9-NEXT: .LBB2_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v12i32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_dual_add_f32 v11, 1.0, v11 :: v_dual_add_f32 v10, 1.0, v10
; GFX11-NEXT: v_dual_add_f32 v9, 1.0, v9 :: v_dual_add_f32 v8, 1.0, v8
; GFX11-NEXT: v_dual_add_f32 v7, 1.0, v7 :: v_dual_add_f32 v6, 1.0, v6
; GFX11-NEXT: v_dual_add_f32 v5, 1.0, v5 :: v_dual_add_f32 v4, 1.0, v4
; GFX11-NEXT: v_dual_add_f32 v3, 1.0, v3 :: v_dual_add_f32 v2, 1.0, v2
; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v0, 1.0, v0
; GFX11-NEXT: ; %bb.2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define inreg <12 x i32> @bitcast_v12f32_to_v12i32_scalar(<12 x float> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v12i32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB3_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB3_2: ; %cmp.true
; SI-NEXT: v_add_f32_e64 v11, s27, 1.0
; SI-NEXT: v_add_f32_e64 v10, s26, 1.0
; SI-NEXT: v_add_f32_e64 v9, s25, 1.0
; SI-NEXT: v_add_f32_e64 v8, s24, 1.0
; SI-NEXT: v_add_f32_e64 v7, s23, 1.0
; SI-NEXT: v_add_f32_e64 v6, s22, 1.0
; SI-NEXT: v_add_f32_e64 v5, s21, 1.0
; SI-NEXT: v_add_f32_e64 v4, s20, 1.0
; SI-NEXT: v_add_f32_e64 v3, s19, 1.0
; SI-NEXT: v_add_f32_e64 v2, s18, 1.0
; SI-NEXT: v_add_f32_e64 v1, s17, 1.0
; SI-NEXT: v_add_f32_e64 v0, s16, 1.0
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v12i32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB3_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB3_2: ; %cmp.true
; VI-NEXT: v_add_f32_e64 v11, s27, 1.0
; VI-NEXT: v_add_f32_e64 v10, s26, 1.0
; VI-NEXT: v_add_f32_e64 v9, s25, 1.0
; VI-NEXT: v_add_f32_e64 v8, s24, 1.0
; VI-NEXT: v_add_f32_e64 v7, s23, 1.0
; VI-NEXT: v_add_f32_e64 v6, s22, 1.0
; VI-NEXT: v_add_f32_e64 v5, s21, 1.0
; VI-NEXT: v_add_f32_e64 v4, s20, 1.0
; VI-NEXT: v_add_f32_e64 v3, s19, 1.0
; VI-NEXT: v_add_f32_e64 v2, s18, 1.0
; VI-NEXT: v_add_f32_e64 v1, s17, 1.0
; VI-NEXT: v_add_f32_e64 v0, s16, 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v12i32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB3_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB3_2: ; %cmp.true
; GFX9-NEXT: v_add_f32_e64 v11, s27, 1.0
; GFX9-NEXT: v_add_f32_e64 v10, s26, 1.0
; GFX9-NEXT: v_add_f32_e64 v9, s25, 1.0
; GFX9-NEXT: v_add_f32_e64 v8, s24, 1.0
; GFX9-NEXT: v_add_f32_e64 v7, s23, 1.0
; GFX9-NEXT: v_add_f32_e64 v6, s22, 1.0
; GFX9-NEXT: v_add_f32_e64 v5, s21, 1.0
; GFX9-NEXT: v_add_f32_e64 v4, s20, 1.0
; GFX9-NEXT: v_add_f32_e64 v3, s19, 1.0
; GFX9-NEXT: v_add_f32_e64 v2, s18, 1.0
; GFX9-NEXT: v_add_f32_e64 v1, s17, 1.0
; GFX9-NEXT: v_add_f32_e64 v0, s16, 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v12i32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB3_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB3_2: ; %cmp.true
; GFX11-NEXT: v_add_f32_e64 v11, s23, 1.0
; GFX11-NEXT: v_add_f32_e64 v10, s22, 1.0
; GFX11-NEXT: v_add_f32_e64 v9, s21, 1.0
; GFX11-NEXT: v_add_f32_e64 v8, s20, 1.0
; GFX11-NEXT: v_add_f32_e64 v7, s19, 1.0
; GFX11-NEXT: v_add_f32_e64 v6, s18, 1.0
; GFX11-NEXT: v_add_f32_e64 v5, s17, 1.0
; GFX11-NEXT: v_add_f32_e64 v4, s16, 1.0
; GFX11-NEXT: v_add_f32_e64 v3, s15, 1.0
; GFX11-NEXT: v_add_f32_e64 v2, s14, 1.0
; GFX11-NEXT: v_add_f32_e64 v1, s13, 1.0
; GFX11-NEXT: v_add_f32_e64 v0, s12, 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define <6 x double> @bitcast_v12i32_to_v6f64(<12 x i32> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v6f64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB4_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v11
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v9
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v7
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v5
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v3
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v1
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: .LBB4_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v6f64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB4_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v11, vcc, 3, v11
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_add_u32_e32 v9, vcc, 3, v9
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_add_u32_e32 v7, vcc, 3, v7
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_add_u32_e32 v5, vcc, 3, v5
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_add_u32_e32 v3, vcc, 3, v3
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_add_u32_e32 v1, vcc, 3, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: .LBB4_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v6f64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB4_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_u32_e32 v11, 3, v11
; GFX9-NEXT: v_add_u32_e32 v10, 3, v10
; GFX9-NEXT: v_add_u32_e32 v9, 3, v9
; GFX9-NEXT: v_add_u32_e32 v8, 3, v8
; GFX9-NEXT: v_add_u32_e32 v7, 3, v7
; GFX9-NEXT: v_add_u32_e32 v6, 3, v6
; GFX9-NEXT: v_add_u32_e32 v5, 3, v5
; GFX9-NEXT: v_add_u32_e32 v4, 3, v4
; GFX9-NEXT: v_add_u32_e32 v3, 3, v3
; GFX9-NEXT: v_add_u32_e32 v2, 3, v2
; GFX9-NEXT: v_add_u32_e32 v1, 3, v1
; GFX9-NEXT: v_add_u32_e32 v0, 3, v0
; GFX9-NEXT: .LBB4_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v6f64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB4_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_nc_u32_e32 v11, 3, v11
; GFX11-NEXT: v_add_nc_u32_e32 v10, 3, v10
; GFX11-NEXT: v_add_nc_u32_e32 v9, 3, v9
; GFX11-NEXT: v_add_nc_u32_e32 v8, 3, v8
; GFX11-NEXT: v_add_nc_u32_e32 v7, 3, v7
; GFX11-NEXT: v_add_nc_u32_e32 v6, 3, v6
; GFX11-NEXT: v_add_nc_u32_e32 v5, 3, v5
; GFX11-NEXT: v_add_nc_u32_e32 v4, 3, v4
; GFX11-NEXT: v_add_nc_u32_e32 v3, 3, v3
; GFX11-NEXT: v_add_nc_u32_e32 v2, 3, v2
; GFX11-NEXT: v_add_nc_u32_e32 v1, 3, v1
; GFX11-NEXT: v_add_nc_u32_e32 v0, 3, v0
; GFX11-NEXT: .LBB4_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define inreg <6 x double> @bitcast_v12i32_to_v6f64_scalar(<12 x i32> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v6f64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc1 .LBB5_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: .LBB5_2: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v6f64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB5_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s27, s27, 3
; VI-NEXT: s_add_i32 s26, s26, 3
; VI-NEXT: s_add_i32 s25, s25, 3
; VI-NEXT: s_add_i32 s24, s24, 3
; VI-NEXT: s_add_i32 s23, s23, 3
; VI-NEXT: s_add_i32 s22, s22, 3
; VI-NEXT: s_add_i32 s21, s21, 3
; VI-NEXT: s_add_i32 s20, s20, 3
; VI-NEXT: s_add_i32 s19, s19, 3
; VI-NEXT: s_add_i32 s18, s18, 3
; VI-NEXT: s_add_i32 s17, s17, 3
; VI-NEXT: s_add_i32 s16, s16, 3
; VI-NEXT: .LBB5_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v6f64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB5_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_i32 s27, s27, 3
; GFX9-NEXT: s_add_i32 s26, s26, 3
; GFX9-NEXT: s_add_i32 s25, s25, 3
; GFX9-NEXT: s_add_i32 s24, s24, 3
; GFX9-NEXT: s_add_i32 s23, s23, 3
; GFX9-NEXT: s_add_i32 s22, s22, 3
; GFX9-NEXT: s_add_i32 s21, s21, 3
; GFX9-NEXT: s_add_i32 s20, s20, 3
; GFX9-NEXT: s_add_i32 s19, s19, 3
; GFX9-NEXT: s_add_i32 s18, s18, 3
; GFX9-NEXT: s_add_i32 s17, s17, 3
; GFX9-NEXT: s_add_i32 s16, s16, 3
; GFX9-NEXT: .LBB5_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v6f64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB5_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_i32 s23, s23, 3
; GFX11-NEXT: s_add_i32 s22, s22, 3
; GFX11-NEXT: s_add_i32 s21, s21, 3
; GFX11-NEXT: s_add_i32 s20, s20, 3
; GFX11-NEXT: s_add_i32 s19, s19, 3
; GFX11-NEXT: s_add_i32 s18, s18, 3
; GFX11-NEXT: s_add_i32 s17, s17, 3
; GFX11-NEXT: s_add_i32 s16, s16, 3
; GFX11-NEXT: s_add_i32 s3, s3, 3
; GFX11-NEXT: s_add_i32 s2, s2, 3
; GFX11-NEXT: s_add_i32 s1, s1, 3
; GFX11-NEXT: s_add_i32 s0, s0, 3
; GFX11-NEXT: .LBB5_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define <12 x i32> @bitcast_v6f64_to_v12i32(<6 x double> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v12i32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB6_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; SI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; SI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; SI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; SI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; SI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; SI-NEXT: .LBB6_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v12i32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB6_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; VI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; VI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; VI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; VI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; VI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; VI-NEXT: .LBB6_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v12i32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB6_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX9-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX9-NEXT: .LBB6_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v12i32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB6_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX11-NEXT: .LBB6_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define inreg <12 x i32> @bitcast_v6f64_to_v12i32_scalar(<6 x double> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v12i32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB7_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB7_2: ; %cmp.true
; SI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; SI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; SI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; SI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; SI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; SI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v12i32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB7_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB7_2: ; %cmp.true
; VI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; VI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; VI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; VI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; VI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; VI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v12i32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB7_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB7_2: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; GFX9-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v12i32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB7_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB7_2: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[10:11], s[22:23], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], s[20:21], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], s[18:19], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], s[16:17], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], s[14:15], 1.0
; GFX11-NEXT: v_add_f64 v[0:1], s[12:13], 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define <6 x i64> @bitcast_v12i32_to_v6i64(<12 x i32> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v6i64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB8_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v11
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v9
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v7
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v5
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v3
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v1
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: .LBB8_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v6i64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB8_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v11, vcc, 3, v11
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_add_u32_e32 v9, vcc, 3, v9
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_add_u32_e32 v7, vcc, 3, v7
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_add_u32_e32 v5, vcc, 3, v5
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_add_u32_e32 v3, vcc, 3, v3
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_add_u32_e32 v1, vcc, 3, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: .LBB8_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v6i64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB8_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_u32_e32 v11, 3, v11
; GFX9-NEXT: v_add_u32_e32 v10, 3, v10
; GFX9-NEXT: v_add_u32_e32 v9, 3, v9
; GFX9-NEXT: v_add_u32_e32 v8, 3, v8
; GFX9-NEXT: v_add_u32_e32 v7, 3, v7
; GFX9-NEXT: v_add_u32_e32 v6, 3, v6
; GFX9-NEXT: v_add_u32_e32 v5, 3, v5
; GFX9-NEXT: v_add_u32_e32 v4, 3, v4
; GFX9-NEXT: v_add_u32_e32 v3, 3, v3
; GFX9-NEXT: v_add_u32_e32 v2, 3, v2
; GFX9-NEXT: v_add_u32_e32 v1, 3, v1
; GFX9-NEXT: v_add_u32_e32 v0, 3, v0
; GFX9-NEXT: .LBB8_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v6i64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB8_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_nc_u32_e32 v11, 3, v11
; GFX11-NEXT: v_add_nc_u32_e32 v10, 3, v10
; GFX11-NEXT: v_add_nc_u32_e32 v9, 3, v9
; GFX11-NEXT: v_add_nc_u32_e32 v8, 3, v8
; GFX11-NEXT: v_add_nc_u32_e32 v7, 3, v7
; GFX11-NEXT: v_add_nc_u32_e32 v6, 3, v6
; GFX11-NEXT: v_add_nc_u32_e32 v5, 3, v5
; GFX11-NEXT: v_add_nc_u32_e32 v4, 3, v4
; GFX11-NEXT: v_add_nc_u32_e32 v3, 3, v3
; GFX11-NEXT: v_add_nc_u32_e32 v2, 3, v2
; GFX11-NEXT: v_add_nc_u32_e32 v1, 3, v1
; GFX11-NEXT: v_add_nc_u32_e32 v0, 3, v0
; GFX11-NEXT: .LBB8_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define inreg <6 x i64> @bitcast_v12i32_to_v6i64_scalar(<12 x i32> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v6i64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc1 .LBB9_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: .LBB9_2: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v6i64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB9_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s27, s27, 3
; VI-NEXT: s_add_i32 s26, s26, 3
; VI-NEXT: s_add_i32 s25, s25, 3
; VI-NEXT: s_add_i32 s24, s24, 3
; VI-NEXT: s_add_i32 s23, s23, 3
; VI-NEXT: s_add_i32 s22, s22, 3
; VI-NEXT: s_add_i32 s21, s21, 3
; VI-NEXT: s_add_i32 s20, s20, 3
; VI-NEXT: s_add_i32 s19, s19, 3
; VI-NEXT: s_add_i32 s18, s18, 3
; VI-NEXT: s_add_i32 s17, s17, 3
; VI-NEXT: s_add_i32 s16, s16, 3
; VI-NEXT: .LBB9_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v6i64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB9_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_i32 s27, s27, 3
; GFX9-NEXT: s_add_i32 s26, s26, 3
; GFX9-NEXT: s_add_i32 s25, s25, 3
; GFX9-NEXT: s_add_i32 s24, s24, 3
; GFX9-NEXT: s_add_i32 s23, s23, 3
; GFX9-NEXT: s_add_i32 s22, s22, 3
; GFX9-NEXT: s_add_i32 s21, s21, 3
; GFX9-NEXT: s_add_i32 s20, s20, 3
; GFX9-NEXT: s_add_i32 s19, s19, 3
; GFX9-NEXT: s_add_i32 s18, s18, 3
; GFX9-NEXT: s_add_i32 s17, s17, 3
; GFX9-NEXT: s_add_i32 s16, s16, 3
; GFX9-NEXT: .LBB9_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v6i64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB9_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_i32 s23, s23, 3
; GFX11-NEXT: s_add_i32 s22, s22, 3
; GFX11-NEXT: s_add_i32 s21, s21, 3
; GFX11-NEXT: s_add_i32 s20, s20, 3
; GFX11-NEXT: s_add_i32 s19, s19, 3
; GFX11-NEXT: s_add_i32 s18, s18, 3
; GFX11-NEXT: s_add_i32 s17, s17, 3
; GFX11-NEXT: s_add_i32 s16, s16, 3
; GFX11-NEXT: s_add_i32 s3, s3, 3
; GFX11-NEXT: s_add_i32 s2, s2, 3
; GFX11-NEXT: s_add_i32 s1, s1, 3
; GFX11-NEXT: s_add_i32 s0, s0, 3
; GFX11-NEXT: .LBB9_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define <12 x i32> @bitcast_v6i64_to_v12i32(<6 x i64> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v12i32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB10_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; SI-NEXT: .LBB10_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v12i32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB10_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: .LBB10_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v12i32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB10_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_co_u32_e32 v10, vcc, 3, v10
; GFX9-NEXT: v_addc_co_u32_e32 v11, vcc, 0, v11, vcc
; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, 3, v8
; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, 0, v9, vcc
; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, 3, v6
; GFX9-NEXT: v_addc_co_u32_e32 v7, vcc, 0, v7, vcc
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, 3, v4
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, 3, v2
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, 3, v0
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: .LBB10_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v12i32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB10_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_co_u32 v10, vcc_lo, v10, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v11, null, 0, v11, vcc_lo
; GFX11-NEXT: v_add_co_u32 v8, vcc_lo, v8, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v9, null, 0, v9, vcc_lo
; GFX11-NEXT: v_add_co_u32 v6, vcc_lo, v6, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v7, null, 0, v7, vcc_lo
; GFX11-NEXT: v_add_co_u32 v4, vcc_lo, v4, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v5, null, 0, v5, vcc_lo
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: .LBB10_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define inreg <12 x i32> @bitcast_v6i64_to_v12i32_scalar(<6 x i64> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v12i32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc1 .LBB11_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: s_add_u32 s26, s26, 3
; SI-NEXT: s_addc_u32 s27, s27, 0
; SI-NEXT: s_add_u32 s24, s24, 3
; SI-NEXT: s_addc_u32 s25, s25, 0
; SI-NEXT: s_add_u32 s22, s22, 3
; SI-NEXT: s_addc_u32 s23, s23, 0
; SI-NEXT: s_add_u32 s20, s20, 3
; SI-NEXT: s_addc_u32 s21, s21, 0
; SI-NEXT: s_add_u32 s18, s18, 3
; SI-NEXT: s_addc_u32 s19, s19, 0
; SI-NEXT: s_add_u32 s16, s16, 3
; SI-NEXT: s_addc_u32 s17, s17, 0
; SI-NEXT: .LBB11_2: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v12i32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB11_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_u32 s26, s26, 3
; VI-NEXT: s_addc_u32 s27, s27, 0
; VI-NEXT: s_add_u32 s24, s24, 3
; VI-NEXT: s_addc_u32 s25, s25, 0
; VI-NEXT: s_add_u32 s22, s22, 3
; VI-NEXT: s_addc_u32 s23, s23, 0
; VI-NEXT: s_add_u32 s20, s20, 3
; VI-NEXT: s_addc_u32 s21, s21, 0
; VI-NEXT: s_add_u32 s18, s18, 3
; VI-NEXT: s_addc_u32 s19, s19, 0
; VI-NEXT: s_add_u32 s16, s16, 3
; VI-NEXT: s_addc_u32 s17, s17, 0
; VI-NEXT: .LBB11_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v12i32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB11_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_u32 s26, s26, 3
; GFX9-NEXT: s_addc_u32 s27, s27, 0
; GFX9-NEXT: s_add_u32 s24, s24, 3
; GFX9-NEXT: s_addc_u32 s25, s25, 0
; GFX9-NEXT: s_add_u32 s22, s22, 3
; GFX9-NEXT: s_addc_u32 s23, s23, 0
; GFX9-NEXT: s_add_u32 s20, s20, 3
; GFX9-NEXT: s_addc_u32 s21, s21, 0
; GFX9-NEXT: s_add_u32 s18, s18, 3
; GFX9-NEXT: s_addc_u32 s19, s19, 0
; GFX9-NEXT: s_add_u32 s16, s16, 3
; GFX9-NEXT: s_addc_u32 s17, s17, 0
; GFX9-NEXT: .LBB11_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v12i32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB11_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_u32 s22, s22, 3
; GFX11-NEXT: s_addc_u32 s23, s23, 0
; GFX11-NEXT: s_add_u32 s20, s20, 3
; GFX11-NEXT: s_addc_u32 s21, s21, 0
; GFX11-NEXT: s_add_u32 s18, s18, 3
; GFX11-NEXT: s_addc_u32 s19, s19, 0
; GFX11-NEXT: s_add_u32 s16, s16, 3
; GFX11-NEXT: s_addc_u32 s17, s17, 0
; GFX11-NEXT: s_add_u32 s2, s2, 3
; GFX11-NEXT: s_addc_u32 s3, s3, 0
; GFX11-NEXT: s_add_u32 s0, s0, 3
; GFX11-NEXT: s_addc_u32 s1, s1, 0
; GFX11-NEXT: .LBB11_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define <24 x i16> @bitcast_v12i32_to_v24i16(<12 x i32> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v24i16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB12_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB12_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB12_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v1
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v3
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v5
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v7
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v9
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v11
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB12_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v0, v0, v20
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v17
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v18
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v17
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v24i16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB12_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v11, vcc, 3, v11
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_add_u32_e32 v9, vcc, 3, v9
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_add_u32_e32 v7, vcc, 3, v7
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_add_u32_e32 v5, vcc, 3, v5
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_add_u32_e32 v3, vcc, 3, v3
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_add_u32_e32 v1, vcc, 3, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: .LBB12_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v24i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB12_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_u32_e32 v11, 3, v11
; GFX9-NEXT: v_add_u32_e32 v10, 3, v10
; GFX9-NEXT: v_add_u32_e32 v9, 3, v9
; GFX9-NEXT: v_add_u32_e32 v8, 3, v8
; GFX9-NEXT: v_add_u32_e32 v7, 3, v7
; GFX9-NEXT: v_add_u32_e32 v6, 3, v6
; GFX9-NEXT: v_add_u32_e32 v5, 3, v5
; GFX9-NEXT: v_add_u32_e32 v4, 3, v4
; GFX9-NEXT: v_add_u32_e32 v3, 3, v3
; GFX9-NEXT: v_add_u32_e32 v2, 3, v2
; GFX9-NEXT: v_add_u32_e32 v1, 3, v1
; GFX9-NEXT: v_add_u32_e32 v0, 3, v0
; GFX9-NEXT: .LBB12_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v24i16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB12_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_nc_u32_e32 v11, 3, v11
; GFX11-NEXT: v_add_nc_u32_e32 v10, 3, v10
; GFX11-NEXT: v_add_nc_u32_e32 v9, 3, v9
; GFX11-NEXT: v_add_nc_u32_e32 v8, 3, v8
; GFX11-NEXT: v_add_nc_u32_e32 v7, 3, v7
; GFX11-NEXT: v_add_nc_u32_e32 v6, 3, v6
; GFX11-NEXT: v_add_nc_u32_e32 v5, 3, v5
; GFX11-NEXT: v_add_nc_u32_e32 v4, 3, v4
; GFX11-NEXT: v_add_nc_u32_e32 v3, 3, v3
; GFX11-NEXT: v_add_nc_u32_e32 v2, 3, v2
; GFX11-NEXT: v_add_nc_u32_e32 v1, 3, v1
; GFX11-NEXT: v_add_nc_u32_e32 v0, 3, v0
; GFX11-NEXT: .LBB12_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define inreg <24 x i16> @bitcast_v12i32_to_v24i16_scalar(<12 x i32> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v24i16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB13_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: s_branch .LBB13_3
; SI-NEXT: .LBB13_2: ; %cmp.true
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: .LBB13_3: ; %end
; SI-NEXT: s_and_b32 s5, s16, 0xffff
; SI-NEXT: s_lshl_b32 s7, s14, 16
; SI-NEXT: s_or_b32 s5, s5, s7
; SI-NEXT: s_and_b32 s7, s17, 0xffff
; SI-NEXT: s_lshl_b32 s9, s43, 16
; SI-NEXT: s_or_b32 s7, s7, s9
; SI-NEXT: s_and_b32 s9, s18, 0xffff
; SI-NEXT: s_lshl_b32 s11, s12, 16
; SI-NEXT: s_or_b32 s9, s9, s11
; SI-NEXT: s_and_b32 s11, s19, 0xffff
; SI-NEXT: s_lshl_b32 s12, s42, 16
; SI-NEXT: s_or_b32 s11, s11, s12
; SI-NEXT: s_and_b32 s12, s20, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s12, s10
; SI-NEXT: s_and_b32 s12, s21, 0xffff
; SI-NEXT: s_lshl_b32 s13, s41, 16
; SI-NEXT: s_or_b32 s12, s12, s13
; SI-NEXT: s_and_b32 s13, s22, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s13, s8
; SI-NEXT: s_and_b32 s13, s23, 0xffff
; SI-NEXT: s_lshl_b32 s14, s40, 16
; SI-NEXT: s_or_b32 s13, s13, s14
; SI-NEXT: s_and_b32 s14, s24, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s14, s6
; SI-NEXT: s_and_b32 s14, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s29, 16
; SI-NEXT: s_or_b32 s14, s14, s15
; SI-NEXT: s_and_b32 s15, s26, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s15, s4
; SI-NEXT: s_and_b32 s15, s27, 0xffff
; SI-NEXT: s_lshl_b32 s16, s28, 16
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: v_mov_b32_e32 v0, s5
; SI-NEXT: v_mov_b32_e32 v1, s7
; SI-NEXT: v_mov_b32_e32 v2, s9
; SI-NEXT: v_mov_b32_e32 v3, s11
; SI-NEXT: v_mov_b32_e32 v4, s10
; SI-NEXT: v_mov_b32_e32 v5, s12
; SI-NEXT: v_mov_b32_e32 v6, s8
; SI-NEXT: v_mov_b32_e32 v7, s13
; SI-NEXT: v_mov_b32_e32 v8, s6
; SI-NEXT: v_mov_b32_e32 v9, s14
; SI-NEXT: v_mov_b32_e32 v10, s4
; SI-NEXT: v_mov_b32_e32 v11, s15
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v24i16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB13_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s27, s27, 3
; VI-NEXT: s_add_i32 s26, s26, 3
; VI-NEXT: s_add_i32 s25, s25, 3
; VI-NEXT: s_add_i32 s24, s24, 3
; VI-NEXT: s_add_i32 s23, s23, 3
; VI-NEXT: s_add_i32 s22, s22, 3
; VI-NEXT: s_add_i32 s21, s21, 3
; VI-NEXT: s_add_i32 s20, s20, 3
; VI-NEXT: s_add_i32 s19, s19, 3
; VI-NEXT: s_add_i32 s18, s18, 3
; VI-NEXT: s_add_i32 s17, s17, 3
; VI-NEXT: s_add_i32 s16, s16, 3
; VI-NEXT: .LBB13_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v24i16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB13_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_i32 s27, s27, 3
; GFX9-NEXT: s_add_i32 s26, s26, 3
; GFX9-NEXT: s_add_i32 s25, s25, 3
; GFX9-NEXT: s_add_i32 s24, s24, 3
; GFX9-NEXT: s_add_i32 s23, s23, 3
; GFX9-NEXT: s_add_i32 s22, s22, 3
; GFX9-NEXT: s_add_i32 s21, s21, 3
; GFX9-NEXT: s_add_i32 s20, s20, 3
; GFX9-NEXT: s_add_i32 s19, s19, 3
; GFX9-NEXT: s_add_i32 s18, s18, 3
; GFX9-NEXT: s_add_i32 s17, s17, 3
; GFX9-NEXT: s_add_i32 s16, s16, 3
; GFX9-NEXT: .LBB13_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v24i16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB13_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_i32 s23, s23, 3
; GFX11-NEXT: s_add_i32 s22, s22, 3
; GFX11-NEXT: s_add_i32 s21, s21, 3
; GFX11-NEXT: s_add_i32 s20, s20, 3
; GFX11-NEXT: s_add_i32 s19, s19, 3
; GFX11-NEXT: s_add_i32 s18, s18, 3
; GFX11-NEXT: s_add_i32 s17, s17, 3
; GFX11-NEXT: s_add_i32 s16, s16, 3
; GFX11-NEXT: s_add_i32 s3, s3, 3
; GFX11-NEXT: s_add_i32 s2, s2, 3
; GFX11-NEXT: s_add_i32 s1, s1, 3
; GFX11-NEXT: s_add_i32 s0, s0, 3
; GFX11-NEXT: .LBB13_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define <12 x i32> @bitcast_v24i16_to_v12i32(<24 x i16> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v12i32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_mov_b32_e32 v24, v11
; SI-NEXT: v_mov_b32_e32 v13, v10
; SI-NEXT: v_mov_b32_e32 v14, v9
; SI-NEXT: v_mov_b32_e32 v15, v8
; SI-NEXT: v_mov_b32_e32 v16, v7
; SI-NEXT: v_mov_b32_e32 v17, v6
; SI-NEXT: v_mov_b32_e32 v18, v5
; SI-NEXT: v_mov_b32_e32 v19, v4
; SI-NEXT: v_mov_b32_e32 v20, v3
; SI-NEXT: v_mov_b32_e32 v21, v2
; SI-NEXT: v_mov_b32_e32 v22, v1
; SI-NEXT: v_mov_b32_e32 v23, v0
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v24
; SI-NEXT: v_lshrrev_b32_e32 v1, 16, v13
; SI-NEXT: v_lshrrev_b32_e32 v2, 16, v14
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v15
; SI-NEXT: v_lshrrev_b32_e32 v4, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v5, 16, v17
; SI-NEXT: v_lshrrev_b32_e32 v6, 16, v18
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v8, 16, v20
; SI-NEXT: v_lshrrev_b32_e32 v9, 16, v21
; SI-NEXT: v_lshrrev_b32_e32 v10, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v11, 16, v23
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: v_lshlrev_b32_e32 v35, 16, v11
; SI-NEXT: v_lshlrev_b32_e32 v34, 16, v10
; SI-NEXT: v_lshlrev_b32_e32 v33, 16, v9
; SI-NEXT: v_lshlrev_b32_e32 v32, 16, v8
; SI-NEXT: v_lshlrev_b32_e32 v31, 16, v7
; SI-NEXT: v_lshlrev_b32_e32 v30, 16, v6
; SI-NEXT: v_lshlrev_b32_e32 v29, 16, v5
; SI-NEXT: v_lshlrev_b32_e32 v28, 16, v4
; SI-NEXT: v_lshlrev_b32_e32 v27, 16, v3
; SI-NEXT: v_lshlrev_b32_e32 v26, 16, v2
; SI-NEXT: v_lshlrev_b32_e32 v25, 16, v1
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v0
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB14_3
; SI-NEXT: ; %bb.1: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB14_4
; SI-NEXT: .LBB14_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB14_3: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v23
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v22
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v21
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v20
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v19
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v18
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v17
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v16
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v15
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v14
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v13
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v24
; SI-NEXT: v_or_b32_e32 v0, v0, v35
; SI-NEXT: v_or_b32_e32 v1, v1, v34
; SI-NEXT: v_or_b32_e32 v2, v2, v33
; SI-NEXT: v_or_b32_e32 v3, v3, v32
; SI-NEXT: v_or_b32_e32 v4, v4, v31
; SI-NEXT: v_or_b32_e32 v5, v5, v30
; SI-NEXT: v_or_b32_e32 v6, v6, v29
; SI-NEXT: v_or_b32_e32 v7, v7, v28
; SI-NEXT: v_or_b32_e32 v8, v8, v27
; SI-NEXT: v_or_b32_e32 v9, v9, v26
; SI-NEXT: v_or_b32_e32 v10, v10, v25
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB14_2
; SI-NEXT: .LBB14_4: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v23
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v22
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v21
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v20
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v19
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v18
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v17
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v16
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v15
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v14
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v13
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v24
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_or_b32_e32 v0, v35, v0
; SI-NEXT: v_or_b32_e32 v1, v34, v1
; SI-NEXT: v_or_b32_e32 v2, v33, v2
; SI-NEXT: v_or_b32_e32 v3, v32, v3
; SI-NEXT: v_or_b32_e32 v4, v31, v4
; SI-NEXT: v_or_b32_e32 v5, v30, v5
; SI-NEXT: v_or_b32_e32 v6, v29, v6
; SI-NEXT: v_or_b32_e32 v7, v28, v7
; SI-NEXT: v_or_b32_e32 v8, v27, v8
; SI-NEXT: v_or_b32_e32 v9, v26, v9
; SI-NEXT: v_or_b32_e32 v10, v25, v10
; SI-NEXT: v_or_b32_e32 v11, v12, v11
; SI-NEXT: v_add_i32_e32 v0, vcc, 0x30000, v0
; SI-NEXT: v_add_i32_e32 v1, vcc, 0x30000, v1
; SI-NEXT: v_add_i32_e32 v2, vcc, 0x30000, v2
; SI-NEXT: v_add_i32_e32 v3, vcc, 0x30000, v3
; SI-NEXT: v_add_i32_e32 v4, vcc, 0x30000, v4
; SI-NEXT: v_add_i32_e32 v5, vcc, 0x30000, v5
; SI-NEXT: v_add_i32_e32 v6, vcc, 0x30000, v6
; SI-NEXT: v_add_i32_e32 v7, vcc, 0x30000, v7
; SI-NEXT: v_add_i32_e32 v8, vcc, 0x30000, v8
; SI-NEXT: v_add_i32_e32 v9, vcc, 0x30000, v9
; SI-NEXT: v_add_i32_e32 v10, vcc, 0x30000, v10
; SI-NEXT: v_add_i32_e32 v11, vcc, 0x30000, v11
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v12i32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB14_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v13, 3
; VI-NEXT: v_add_u16_e32 v12, 3, v11
; VI-NEXT: v_add_u16_sdwa v11, v11, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v11, v12, v11
; VI-NEXT: v_add_u16_e32 v12, 3, v10
; VI-NEXT: v_add_u16_sdwa v10, v10, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v10, v12, v10
; VI-NEXT: v_add_u16_e32 v12, 3, v9
; VI-NEXT: v_add_u16_sdwa v9, v9, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v9, v12, v9
; VI-NEXT: v_add_u16_e32 v12, 3, v8
; VI-NEXT: v_add_u16_sdwa v8, v8, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v8, v12, v8
; VI-NEXT: v_add_u16_e32 v12, 3, v7
; VI-NEXT: v_add_u16_sdwa v7, v7, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v7, v12, v7
; VI-NEXT: v_add_u16_e32 v12, 3, v6
; VI-NEXT: v_add_u16_sdwa v6, v6, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v6, v12, v6
; VI-NEXT: v_add_u16_e32 v12, 3, v5
; VI-NEXT: v_add_u16_sdwa v5, v5, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v5, v12, v5
; VI-NEXT: v_add_u16_e32 v12, 3, v4
; VI-NEXT: v_add_u16_sdwa v4, v4, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v4, v12, v4
; VI-NEXT: v_add_u16_e32 v12, 3, v3
; VI-NEXT: v_add_u16_sdwa v3, v3, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v3, v12, v3
; VI-NEXT: v_add_u16_e32 v12, 3, v2
; VI-NEXT: v_add_u16_sdwa v2, v2, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v2, v12, v2
; VI-NEXT: v_add_u16_e32 v12, 3, v1
; VI-NEXT: v_add_u16_sdwa v1, v1, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v1, v12, v1
; VI-NEXT: v_add_u16_e32 v12, 3, v0
; VI-NEXT: v_add_u16_sdwa v0, v0, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v0, v12, v0
; VI-NEXT: .LBB14_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v12i32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB14_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB14_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v12i32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB14_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB14_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define inreg <12 x i32> @bitcast_v24i16_to_v12i32_scalar(<24 x i16> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v12i32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s12, s27, 16
; SI-NEXT: s_lshr_b32 s14, s26, 16
; SI-NEXT: s_lshr_b32 s15, s25, 16
; SI-NEXT: s_lshr_b32 s13, s24, 16
; SI-NEXT: s_lshr_b32 s11, s23, 16
; SI-NEXT: s_lshr_b32 s10, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s8, s20, 16
; SI-NEXT: s_lshr_b32 s7, s19, 16
; SI-NEXT: s_lshr_b32 s6, s18, 16
; SI-NEXT: s_lshr_b32 s5, s17, 16
; SI-NEXT: s_lshr_b32 s4, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB15_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s16, s4
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_or_b32 s5, s16, s5
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s16, s6
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_or_b32 s7, s16, s7
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s16, s8
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_or_b32 s9, s16, s9
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s16, s10
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_or_b32 s11, s16, s11
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_or_b32 s13, s16, s13
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_or_b32 s14, s16, s14
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s16, s12
; SI-NEXT: s_branch .LBB15_3
; SI-NEXT: .LBB15_2: ; %cmp.true
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_or_b32 s4, s4, s16
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_or_b32 s5, s5, s16
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_or_b32 s6, s6, s16
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_or_b32 s7, s7, s16
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_or_b32 s8, s8, s16
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_or_b32 s9, s9, s16
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_or_b32 s10, s10, s16
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_or_b32 s11, s11, s16
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_or_b32 s13, s13, s16
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_or_b32 s14, s14, s16
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s12, s16
; SI-NEXT: s_add_i32 s4, s4, 0x30000
; SI-NEXT: s_add_i32 s5, s5, 0x30000
; SI-NEXT: s_add_i32 s6, s6, 0x30000
; SI-NEXT: s_add_i32 s7, s7, 0x30000
; SI-NEXT: s_add_i32 s8, s8, 0x30000
; SI-NEXT: s_add_i32 s9, s9, 0x30000
; SI-NEXT: s_add_i32 s10, s10, 0x30000
; SI-NEXT: s_add_i32 s11, s11, 0x30000
; SI-NEXT: s_add_i32 s13, s13, 0x30000
; SI-NEXT: s_add_i32 s15, s15, 0x30000
; SI-NEXT: s_add_i32 s14, s14, 0x30000
; SI-NEXT: s_add_i32 s12, s12, 0x30000
; SI-NEXT: .LBB15_3: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: v_mov_b32_e32 v2, s6
; SI-NEXT: v_mov_b32_e32 v3, s7
; SI-NEXT: v_mov_b32_e32 v4, s8
; SI-NEXT: v_mov_b32_e32 v5, s9
; SI-NEXT: v_mov_b32_e32 v6, s10
; SI-NEXT: v_mov_b32_e32 v7, s11
; SI-NEXT: v_mov_b32_e32 v8, s13
; SI-NEXT: v_mov_b32_e32 v9, s15
; SI-NEXT: v_mov_b32_e32 v10, s14
; SI-NEXT: v_mov_b32_e32 v11, s12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v12i32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB15_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s5, s27, 3
; VI-NEXT: s_and_b32 s4, s27, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s26, 3
; VI-NEXT: s_add_i32 s27, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s26, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s25, 3
; VI-NEXT: s_add_i32 s26, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s25, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s24, 3
; VI-NEXT: s_add_i32 s25, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s24, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s23, 3
; VI-NEXT: s_add_i32 s24, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s23, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s22, 3
; VI-NEXT: s_add_i32 s23, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s22, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s21, 3
; VI-NEXT: s_add_i32 s22, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s21, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s20, 3
; VI-NEXT: s_add_i32 s21, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s20, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s19, 3
; VI-NEXT: s_add_i32 s20, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s19, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s18, 3
; VI-NEXT: s_add_i32 s19, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s18, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s17, 3
; VI-NEXT: s_add_i32 s18, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s17, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s16, 3
; VI-NEXT: s_add_i32 s17, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s16, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s16, s4, 0x30000
; VI-NEXT: .LBB15_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v12i32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB15_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB15_2: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, s27, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, s26, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, s25, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, s24, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, s23, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, s22, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, s21, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, s20, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, s19, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, s18, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, s17, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, s16, 3 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v12i32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB15_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB15_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, s23, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, s22, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, s21, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, s20, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, s19, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, s18, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, s17, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, s16, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, s15, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, s14, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, s13, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, s12, 3 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define <24 x half> @bitcast_v12i32_to_v24f16(<12 x i32> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v24f16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB16_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB16_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB16_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v1
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v3
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v5
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v7
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v9
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v11
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB16_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v0, v0, v20
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v17
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v18
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v17
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v24f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB16_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v11, vcc, 3, v11
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_add_u32_e32 v9, vcc, 3, v9
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_add_u32_e32 v7, vcc, 3, v7
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_add_u32_e32 v5, vcc, 3, v5
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_add_u32_e32 v3, vcc, 3, v3
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_add_u32_e32 v1, vcc, 3, v1
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: .LBB16_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v24f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB16_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_u32_e32 v11, 3, v11
; GFX9-NEXT: v_add_u32_e32 v10, 3, v10
; GFX9-NEXT: v_add_u32_e32 v9, 3, v9
; GFX9-NEXT: v_add_u32_e32 v8, 3, v8
; GFX9-NEXT: v_add_u32_e32 v7, 3, v7
; GFX9-NEXT: v_add_u32_e32 v6, 3, v6
; GFX9-NEXT: v_add_u32_e32 v5, 3, v5
; GFX9-NEXT: v_add_u32_e32 v4, 3, v4
; GFX9-NEXT: v_add_u32_e32 v3, 3, v3
; GFX9-NEXT: v_add_u32_e32 v2, 3, v2
; GFX9-NEXT: v_add_u32_e32 v1, 3, v1
; GFX9-NEXT: v_add_u32_e32 v0, 3, v0
; GFX9-NEXT: .LBB16_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v24f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB16_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_nc_u32_e32 v11, 3, v11
; GFX11-NEXT: v_add_nc_u32_e32 v10, 3, v10
; GFX11-NEXT: v_add_nc_u32_e32 v9, 3, v9
; GFX11-NEXT: v_add_nc_u32_e32 v8, 3, v8
; GFX11-NEXT: v_add_nc_u32_e32 v7, 3, v7
; GFX11-NEXT: v_add_nc_u32_e32 v6, 3, v6
; GFX11-NEXT: v_add_nc_u32_e32 v5, 3, v5
; GFX11-NEXT: v_add_nc_u32_e32 v4, 3, v4
; GFX11-NEXT: v_add_nc_u32_e32 v3, 3, v3
; GFX11-NEXT: v_add_nc_u32_e32 v2, 3, v2
; GFX11-NEXT: v_add_nc_u32_e32 v1, 3, v1
; GFX11-NEXT: v_add_nc_u32_e32 v0, 3, v0
; GFX11-NEXT: .LBB16_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define inreg <24 x half> @bitcast_v12i32_to_v24f16_scalar(<12 x i32> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12i32_to_v24f16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB17_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: s_branch .LBB17_3
; SI-NEXT: .LBB17_2: ; %cmp.true
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: .LBB17_3: ; %end
; SI-NEXT: s_and_b32 s5, s16, 0xffff
; SI-NEXT: s_lshl_b32 s7, s14, 16
; SI-NEXT: s_or_b32 s5, s5, s7
; SI-NEXT: s_and_b32 s7, s17, 0xffff
; SI-NEXT: s_lshl_b32 s9, s43, 16
; SI-NEXT: s_or_b32 s7, s7, s9
; SI-NEXT: s_and_b32 s9, s18, 0xffff
; SI-NEXT: s_lshl_b32 s11, s12, 16
; SI-NEXT: s_or_b32 s9, s9, s11
; SI-NEXT: s_and_b32 s11, s19, 0xffff
; SI-NEXT: s_lshl_b32 s12, s42, 16
; SI-NEXT: s_or_b32 s11, s11, s12
; SI-NEXT: s_and_b32 s12, s20, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s12, s10
; SI-NEXT: s_and_b32 s12, s21, 0xffff
; SI-NEXT: s_lshl_b32 s13, s41, 16
; SI-NEXT: s_or_b32 s12, s12, s13
; SI-NEXT: s_and_b32 s13, s22, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s13, s8
; SI-NEXT: s_and_b32 s13, s23, 0xffff
; SI-NEXT: s_lshl_b32 s14, s40, 16
; SI-NEXT: s_or_b32 s13, s13, s14
; SI-NEXT: s_and_b32 s14, s24, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s14, s6
; SI-NEXT: s_and_b32 s14, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s29, 16
; SI-NEXT: s_or_b32 s14, s14, s15
; SI-NEXT: s_and_b32 s15, s26, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s15, s4
; SI-NEXT: s_and_b32 s15, s27, 0xffff
; SI-NEXT: s_lshl_b32 s16, s28, 16
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: v_mov_b32_e32 v0, s5
; SI-NEXT: v_mov_b32_e32 v1, s7
; SI-NEXT: v_mov_b32_e32 v2, s9
; SI-NEXT: v_mov_b32_e32 v3, s11
; SI-NEXT: v_mov_b32_e32 v4, s10
; SI-NEXT: v_mov_b32_e32 v5, s12
; SI-NEXT: v_mov_b32_e32 v6, s8
; SI-NEXT: v_mov_b32_e32 v7, s13
; SI-NEXT: v_mov_b32_e32 v8, s6
; SI-NEXT: v_mov_b32_e32 v9, s14
; SI-NEXT: v_mov_b32_e32 v10, s4
; SI-NEXT: v_mov_b32_e32 v11, s15
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12i32_to_v24f16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB17_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s27, s27, 3
; VI-NEXT: s_add_i32 s26, s26, 3
; VI-NEXT: s_add_i32 s25, s25, 3
; VI-NEXT: s_add_i32 s24, s24, 3
; VI-NEXT: s_add_i32 s23, s23, 3
; VI-NEXT: s_add_i32 s22, s22, 3
; VI-NEXT: s_add_i32 s21, s21, 3
; VI-NEXT: s_add_i32 s20, s20, 3
; VI-NEXT: s_add_i32 s19, s19, 3
; VI-NEXT: s_add_i32 s18, s18, 3
; VI-NEXT: s_add_i32 s17, s17, 3
; VI-NEXT: s_add_i32 s16, s16, 3
; VI-NEXT: .LBB17_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12i32_to_v24f16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB17_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_i32 s27, s27, 3
; GFX9-NEXT: s_add_i32 s26, s26, 3
; GFX9-NEXT: s_add_i32 s25, s25, 3
; GFX9-NEXT: s_add_i32 s24, s24, 3
; GFX9-NEXT: s_add_i32 s23, s23, 3
; GFX9-NEXT: s_add_i32 s22, s22, 3
; GFX9-NEXT: s_add_i32 s21, s21, 3
; GFX9-NEXT: s_add_i32 s20, s20, 3
; GFX9-NEXT: s_add_i32 s19, s19, 3
; GFX9-NEXT: s_add_i32 s18, s18, 3
; GFX9-NEXT: s_add_i32 s17, s17, 3
; GFX9-NEXT: s_add_i32 s16, s16, 3
; GFX9-NEXT: .LBB17_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12i32_to_v24f16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB17_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_i32 s23, s23, 3
; GFX11-NEXT: s_add_i32 s22, s22, 3
; GFX11-NEXT: s_add_i32 s21, s21, 3
; GFX11-NEXT: s_add_i32 s20, s20, 3
; GFX11-NEXT: s_add_i32 s19, s19, 3
; GFX11-NEXT: s_add_i32 s18, s18, 3
; GFX11-NEXT: s_add_i32 s17, s17, 3
; GFX11-NEXT: s_add_i32 s16, s16, 3
; GFX11-NEXT: s_add_i32 s3, s3, 3
; GFX11-NEXT: s_add_i32 s2, s2, 3
; GFX11-NEXT: s_add_i32 s1, s1, 3
; GFX11-NEXT: s_add_i32 s0, s0, 3
; GFX11-NEXT: .LBB17_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <12 x i32> %a, splat (i32 3)
%a2 = bitcast <12 x i32> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <12 x i32> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define <12 x i32> @bitcast_v24f16_to_v12i32(<24 x half> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v12i32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_mov_b32_e32 v24, v11
; SI-NEXT: v_mov_b32_e32 v13, v10
; SI-NEXT: v_mov_b32_e32 v14, v9
; SI-NEXT: v_mov_b32_e32 v15, v8
; SI-NEXT: v_mov_b32_e32 v16, v7
; SI-NEXT: v_mov_b32_e32 v17, v6
; SI-NEXT: v_mov_b32_e32 v18, v5
; SI-NEXT: v_mov_b32_e32 v19, v4
; SI-NEXT: v_mov_b32_e32 v20, v3
; SI-NEXT: v_mov_b32_e32 v21, v2
; SI-NEXT: v_mov_b32_e32 v22, v1
; SI-NEXT: v_mov_b32_e32 v23, v0
; SI-NEXT: v_lshrrev_b32_e32 v25, 16, v24
; SI-NEXT: v_lshrrev_b32_e32 v26, 16, v13
; SI-NEXT: v_lshrrev_b32_e32 v27, 16, v14
; SI-NEXT: v_lshrrev_b32_e32 v28, 16, v15
; SI-NEXT: v_lshrrev_b32_e32 v29, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v30, 16, v17
; SI-NEXT: v_lshrrev_b32_e32 v31, 16, v18
; SI-NEXT: v_lshrrev_b32_e32 v32, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v33, 16, v20
; SI-NEXT: v_lshrrev_b32_e32 v34, 16, v21
; SI-NEXT: v_lshrrev_b32_e32 v35, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v36, 16, v23
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB18_3
; SI-NEXT: ; %bb.1: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB18_4
; SI-NEXT: .LBB18_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB18_3: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v23
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v36
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v22
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v35
; SI-NEXT: v_or_b32_e32 v0, v0, v1
; SI-NEXT: v_or_b32_e32 v1, v2, v3
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v21
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v34
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v20
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v33
; SI-NEXT: v_or_b32_e32 v3, v3, v4
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v19
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v32
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v18
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v31
; SI-NEXT: v_or_b32_e32 v5, v5, v6
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v17
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v30
; SI-NEXT: v_or_b32_e32 v6, v6, v7
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v16
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v29
; SI-NEXT: v_or_b32_e32 v7, v7, v8
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v15
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v28
; SI-NEXT: v_or_b32_e32 v8, v8, v9
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v14
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v27
; SI-NEXT: v_or_b32_e32 v9, v9, v10
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v13
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v26
; SI-NEXT: v_or_b32_e32 v10, v10, v11
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v24
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v25
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr36
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB18_2
; SI-NEXT: .LBB18_4: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v0, v36
; SI-NEXT: v_cvt_f32_f16_e32 v2, v35
; SI-NEXT: v_cvt_f32_f16_e32 v1, v23
; SI-NEXT: v_cvt_f32_f16_e32 v3, v22
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT: v_or_b32_e32 v0, v1, v0
; SI-NEXT: v_or_b32_e32 v1, v3, v2
; SI-NEXT: v_cvt_f32_f16_e32 v2, v34
; SI-NEXT: v_cvt_f32_f16_e32 v3, v21
; SI-NEXT: v_cvt_f32_f16_e32 v4, v33
; SI-NEXT: v_cvt_f32_f16_e32 v5, v20
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v6, v32
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT: v_or_b32_e32 v2, v3, v2
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_or_b32_e32 v3, v5, v3
; SI-NEXT: v_cvt_f32_f16_e32 v4, v19
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, v31
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, v18
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, v30
; SI-NEXT: v_cvt_f32_f16_e32 v8, v17
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_or_b32_e32 v5, v7, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, v29
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT: v_or_b32_e32 v6, v8, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, v16
; SI-NEXT: v_cvt_f32_f16_e32 v9, v28
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT: v_cvt_f32_f16_e32 v10, v15
; SI-NEXT: v_or_b32_e32 v7, v8, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v9
; SI-NEXT: v_cvt_f32_f16_e32 v9, v27
; SI-NEXT: v_cvt_f32_f16_e32 v11, v14
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_or_b32_e32 v8, v10, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, v26
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
; SI-NEXT: v_or_b32_e32 v9, v11, v9
; SI-NEXT: v_cvt_f32_f16_e32 v11, v13
; SI-NEXT: v_cvt_f32_f16_e32 v12, v25
; SI-NEXT: v_cvt_f32_f16_e32 v13, v24
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT: v_or_b32_e32 v10, v11, v10
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT: v_or_b32_e32 v11, v13, v11
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v12i32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB18_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v12, 0x200
; VI-NEXT: v_add_f16_sdwa v13, v11, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v11, 0x200, v11
; VI-NEXT: v_or_b32_e32 v11, v11, v13
; VI-NEXT: v_add_f16_sdwa v13, v10, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v10, 0x200, v10
; VI-NEXT: v_or_b32_e32 v10, v10, v13
; VI-NEXT: v_add_f16_sdwa v13, v9, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v9, 0x200, v9
; VI-NEXT: v_or_b32_e32 v9, v9, v13
; VI-NEXT: v_add_f16_sdwa v13, v8, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v8, 0x200, v8
; VI-NEXT: v_or_b32_e32 v8, v8, v13
; VI-NEXT: v_add_f16_sdwa v13, v7, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v7, 0x200, v7
; VI-NEXT: v_or_b32_e32 v7, v7, v13
; VI-NEXT: v_add_f16_sdwa v13, v6, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v6, 0x200, v6
; VI-NEXT: v_or_b32_e32 v6, v6, v13
; VI-NEXT: v_add_f16_sdwa v13, v5, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v5, 0x200, v5
; VI-NEXT: v_or_b32_e32 v5, v5, v13
; VI-NEXT: v_add_f16_sdwa v13, v4, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v4, 0x200, v4
; VI-NEXT: v_or_b32_e32 v4, v4, v13
; VI-NEXT: v_add_f16_sdwa v13, v3, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v3, 0x200, v3
; VI-NEXT: v_or_b32_e32 v3, v3, v13
; VI-NEXT: v_add_f16_sdwa v13, v2, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, 0x200, v2
; VI-NEXT: v_or_b32_e32 v2, v2, v13
; VI-NEXT: v_add_f16_sdwa v13, v1, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v1, 0x200, v1
; VI-NEXT: v_add_f16_sdwa v12, v0, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, 0x200, v0
; VI-NEXT: v_or_b32_e32 v1, v1, v13
; VI-NEXT: v_or_b32_e32 v0, v0, v12
; VI-NEXT: .LBB18_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v12i32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB18_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_movk_i32 s6, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, v11, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, v10, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, v9, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, v8, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, v7, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, v6, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, v5, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, v4, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, v3, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, v2, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, v1, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, v0, s6 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB18_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v12i32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB18_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, v11, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, v10, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, v9, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, v8, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, v7, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, v6, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, v5, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, v4, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, v3, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, v2, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, v1, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, v0, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB18_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define inreg <12 x i32> @bitcast_v24f16_to_v12i32_scalar(<24 x half> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v12i32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s15, s27, 16
; SI-NEXT: s_lshr_b32 s14, s26, 16
; SI-NEXT: s_lshr_b32 s13, s25, 16
; SI-NEXT: s_lshr_b32 s12, s24, 16
; SI-NEXT: s_lshr_b32 s11, s23, 16
; SI-NEXT: s_lshr_b32 s10, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s8, s20, 16
; SI-NEXT: s_lshr_b32 s7, s19, 16
; SI-NEXT: s_lshr_b32 s6, s18, 16
; SI-NEXT: s_lshr_b32 s5, s17, 16
; SI-NEXT: s_lshr_b32 s4, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB19_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s16, s4
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_or_b32 s5, s16, s5
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s16, s6
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_or_b32 s7, s16, s7
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s16, s8
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_or_b32 s9, s16, s9
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s16, s10
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_or_b32 s11, s16, s11
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s16, s12
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_or_b32 s13, s16, s13
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_or_b32 s14, s16, s14
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: v_mov_b32_e32 v2, s6
; SI-NEXT: v_mov_b32_e32 v3, s7
; SI-NEXT: v_mov_b32_e32 v4, s8
; SI-NEXT: v_mov_b32_e32 v5, s9
; SI-NEXT: v_mov_b32_e32 v6, s10
; SI-NEXT: v_mov_b32_e32 v7, s11
; SI-NEXT: v_mov_b32_e32 v8, s12
; SI-NEXT: v_mov_b32_e32 v9, s13
; SI-NEXT: v_mov_b32_e32 v10, s14
; SI-NEXT: v_mov_b32_e32 v11, s15
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB19_2: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v0, s4
; SI-NEXT: v_cvt_f32_f16_e32 v1, s16
; SI-NEXT: v_cvt_f32_f16_e32 v2, s5
; SI-NEXT: v_cvt_f32_f16_e32 v3, s17
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_or_b32_e32 v0, v1, v0
; SI-NEXT: v_cvt_f32_f16_e32 v1, s6
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f32_f16_e32 v5, s19
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v4, v1
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
; SI-NEXT: v_cvt_f32_f16_e32 v2, s18
; SI-NEXT: v_or_b32_e32 v1, v3, v1
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_cvt_f32_f16_e32 v4, s7
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v6, s8
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_or_b32_e32 v3, v5, v3
; SI-NEXT: v_cvt_f32_f16_e32 v4, s20
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, s9
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, s21
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, s10
; SI-NEXT: v_cvt_f32_f16_e32 v8, s22
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_or_b32_e32 v5, v7, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, s11
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT: v_or_b32_e32 v6, v8, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, s23
; SI-NEXT: v_cvt_f32_f16_e32 v9, s12
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT: v_cvt_f32_f16_e32 v10, s24
; SI-NEXT: v_or_b32_e32 v7, v8, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v9
; SI-NEXT: v_cvt_f32_f16_e32 v9, s13
; SI-NEXT: v_cvt_f32_f16_e32 v11, s25
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_or_b32_e32 v8, v10, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, s14
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
; SI-NEXT: v_or_b32_e32 v9, v11, v9
; SI-NEXT: v_cvt_f32_f16_e32 v11, s26
; SI-NEXT: v_cvt_f32_f16_e32 v12, s15
; SI-NEXT: v_cvt_f32_f16_e32 v13, s27
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT: v_or_b32_e32 v10, v11, v10
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT: v_or_b32_e32 v11, v13, v11
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v12i32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB19_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB19_2: ; %cmp.true
; VI-NEXT: s_lshr_b32 s4, s27, 16
; VI-NEXT: v_mov_b32_e32 v0, 0x200
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s27, v0
; VI-NEXT: s_lshr_b32 s4, s26, 16
; VI-NEXT: v_or_b32_e32 v11, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s26, v0
; VI-NEXT: s_lshr_b32 s4, s25, 16
; VI-NEXT: v_or_b32_e32 v10, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s25, v0
; VI-NEXT: s_lshr_b32 s4, s24, 16
; VI-NEXT: v_or_b32_e32 v9, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s24, v0
; VI-NEXT: s_lshr_b32 s4, s23, 16
; VI-NEXT: v_or_b32_e32 v8, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s23, v0
; VI-NEXT: s_lshr_b32 s4, s22, 16
; VI-NEXT: v_or_b32_e32 v7, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s22, v0
; VI-NEXT: s_lshr_b32 s4, s21, 16
; VI-NEXT: v_or_b32_e32 v6, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s21, v0
; VI-NEXT: s_lshr_b32 s4, s20, 16
; VI-NEXT: v_or_b32_e32 v5, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s20, v0
; VI-NEXT: s_lshr_b32 s4, s19, 16
; VI-NEXT: v_or_b32_e32 v4, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s19, v0
; VI-NEXT: s_lshr_b32 s4, s18, 16
; VI-NEXT: v_or_b32_e32 v3, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s18, v0
; VI-NEXT: s_lshr_b32 s4, s17, 16
; VI-NEXT: v_or_b32_e32 v2, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v12, s17, v0
; VI-NEXT: s_lshr_b32 s4, s16, 16
; VI-NEXT: v_or_b32_e32 v1, v12, v1
; VI-NEXT: v_mov_b32_e32 v12, s4
; VI-NEXT: v_add_f16_sdwa v12, v12, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, s16, v0
; VI-NEXT: v_or_b32_e32 v0, v0, v12
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v12i32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB19_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB19_2: ; %cmp.true
; GFX9-NEXT: v_mov_b32_e32 v0, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, s27, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, s26, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, s25, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, s24, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, s23, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, s22, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, s21, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, s20, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, s19, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, s18, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, s17, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, s16, v0 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v12i32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB19_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB19_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, s23, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, s22, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, s21, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, s20, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, s19, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, s18, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, s17, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, s16, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, s15, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, s14, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, s13, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, s12, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <12 x i32>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <12 x i32>
br label %end
end:
%phi = phi <12 x i32> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x i32> %phi
}
define <6 x double> @bitcast_v12f32_to_v6f64(<12 x float> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v6f64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB20_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_f32_e32 v11, 1.0, v11
; SI-NEXT: v_add_f32_e32 v10, 1.0, v10
; SI-NEXT: v_add_f32_e32 v9, 1.0, v9
; SI-NEXT: v_add_f32_e32 v8, 1.0, v8
; SI-NEXT: v_add_f32_e32 v7, 1.0, v7
; SI-NEXT: v_add_f32_e32 v6, 1.0, v6
; SI-NEXT: v_add_f32_e32 v5, 1.0, v5
; SI-NEXT: v_add_f32_e32 v4, 1.0, v4
; SI-NEXT: v_add_f32_e32 v3, 1.0, v3
; SI-NEXT: v_add_f32_e32 v2, 1.0, v2
; SI-NEXT: v_add_f32_e32 v1, 1.0, v1
; SI-NEXT: v_add_f32_e32 v0, 1.0, v0
; SI-NEXT: .LBB20_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v6f64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB20_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f32_e32 v11, 1.0, v11
; VI-NEXT: v_add_f32_e32 v10, 1.0, v10
; VI-NEXT: v_add_f32_e32 v9, 1.0, v9
; VI-NEXT: v_add_f32_e32 v8, 1.0, v8
; VI-NEXT: v_add_f32_e32 v7, 1.0, v7
; VI-NEXT: v_add_f32_e32 v6, 1.0, v6
; VI-NEXT: v_add_f32_e32 v5, 1.0, v5
; VI-NEXT: v_add_f32_e32 v4, 1.0, v4
; VI-NEXT: v_add_f32_e32 v3, 1.0, v3
; VI-NEXT: v_add_f32_e32 v2, 1.0, v2
; VI-NEXT: v_add_f32_e32 v1, 1.0, v1
; VI-NEXT: v_add_f32_e32 v0, 1.0, v0
; VI-NEXT: .LBB20_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v6f64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB20_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f32_e32 v11, 1.0, v11
; GFX9-NEXT: v_add_f32_e32 v10, 1.0, v10
; GFX9-NEXT: v_add_f32_e32 v9, 1.0, v9
; GFX9-NEXT: v_add_f32_e32 v8, 1.0, v8
; GFX9-NEXT: v_add_f32_e32 v7, 1.0, v7
; GFX9-NEXT: v_add_f32_e32 v6, 1.0, v6
; GFX9-NEXT: v_add_f32_e32 v5, 1.0, v5
; GFX9-NEXT: v_add_f32_e32 v4, 1.0, v4
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v3
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v2
; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
; GFX9-NEXT: v_add_f32_e32 v0, 1.0, v0
; GFX9-NEXT: .LBB20_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v6f64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_dual_add_f32 v11, 1.0, v11 :: v_dual_add_f32 v10, 1.0, v10
; GFX11-NEXT: v_dual_add_f32 v9, 1.0, v9 :: v_dual_add_f32 v8, 1.0, v8
; GFX11-NEXT: v_dual_add_f32 v7, 1.0, v7 :: v_dual_add_f32 v6, 1.0, v6
; GFX11-NEXT: v_dual_add_f32 v5, 1.0, v5 :: v_dual_add_f32 v4, 1.0, v4
; GFX11-NEXT: v_dual_add_f32 v3, 1.0, v3 :: v_dual_add_f32 v2, 1.0, v2
; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v0, 1.0, v0
; GFX11-NEXT: ; %bb.2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define inreg <6 x double> @bitcast_v12f32_to_v6f64_scalar(<12 x float> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v6f64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB21_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: v_mov_b32_e32 v12, s28
; SI-NEXT: v_mov_b32_e32 v13, s29
; SI-NEXT: v_mov_b32_e32 v14, s30
; SI-NEXT: v_mov_b32_e32 v15, s31
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB21_2: ; %cmp.true
; SI-NEXT: v_add_f32_e64 v11, s27, 1.0
; SI-NEXT: v_add_f32_e64 v10, s26, 1.0
; SI-NEXT: v_add_f32_e64 v9, s25, 1.0
; SI-NEXT: v_add_f32_e64 v8, s24, 1.0
; SI-NEXT: v_add_f32_e64 v7, s23, 1.0
; SI-NEXT: v_add_f32_e64 v6, s22, 1.0
; SI-NEXT: v_add_f32_e64 v5, s21, 1.0
; SI-NEXT: v_add_f32_e64 v4, s20, 1.0
; SI-NEXT: v_add_f32_e64 v3, s19, 1.0
; SI-NEXT: v_add_f32_e64 v2, s18, 1.0
; SI-NEXT: v_add_f32_e64 v1, s17, 1.0
; SI-NEXT: v_add_f32_e64 v0, s16, 1.0
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v6f64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB21_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB21_2: ; %cmp.true
; VI-NEXT: v_add_f32_e64 v11, s27, 1.0
; VI-NEXT: v_add_f32_e64 v10, s26, 1.0
; VI-NEXT: v_add_f32_e64 v9, s25, 1.0
; VI-NEXT: v_add_f32_e64 v8, s24, 1.0
; VI-NEXT: v_add_f32_e64 v7, s23, 1.0
; VI-NEXT: v_add_f32_e64 v6, s22, 1.0
; VI-NEXT: v_add_f32_e64 v5, s21, 1.0
; VI-NEXT: v_add_f32_e64 v4, s20, 1.0
; VI-NEXT: v_add_f32_e64 v3, s19, 1.0
; VI-NEXT: v_add_f32_e64 v2, s18, 1.0
; VI-NEXT: v_add_f32_e64 v1, s17, 1.0
; VI-NEXT: v_add_f32_e64 v0, s16, 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v6f64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB21_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB21_2: ; %cmp.true
; GFX9-NEXT: v_add_f32_e64 v11, s27, 1.0
; GFX9-NEXT: v_add_f32_e64 v10, s26, 1.0
; GFX9-NEXT: v_add_f32_e64 v9, s25, 1.0
; GFX9-NEXT: v_add_f32_e64 v8, s24, 1.0
; GFX9-NEXT: v_add_f32_e64 v7, s23, 1.0
; GFX9-NEXT: v_add_f32_e64 v6, s22, 1.0
; GFX9-NEXT: v_add_f32_e64 v5, s21, 1.0
; GFX9-NEXT: v_add_f32_e64 v4, s20, 1.0
; GFX9-NEXT: v_add_f32_e64 v3, s19, 1.0
; GFX9-NEXT: v_add_f32_e64 v2, s18, 1.0
; GFX9-NEXT: v_add_f32_e64 v1, s17, 1.0
; GFX9-NEXT: v_add_f32_e64 v0, s16, 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v6f64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB21_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB21_2: ; %cmp.true
; GFX11-NEXT: v_add_f32_e64 v11, s23, 1.0
; GFX11-NEXT: v_add_f32_e64 v10, s22, 1.0
; GFX11-NEXT: v_add_f32_e64 v9, s21, 1.0
; GFX11-NEXT: v_add_f32_e64 v8, s20, 1.0
; GFX11-NEXT: v_add_f32_e64 v7, s19, 1.0
; GFX11-NEXT: v_add_f32_e64 v6, s18, 1.0
; GFX11-NEXT: v_add_f32_e64 v5, s17, 1.0
; GFX11-NEXT: v_add_f32_e64 v4, s16, 1.0
; GFX11-NEXT: v_add_f32_e64 v3, s15, 1.0
; GFX11-NEXT: v_add_f32_e64 v2, s14, 1.0
; GFX11-NEXT: v_add_f32_e64 v1, s13, 1.0
; GFX11-NEXT: v_add_f32_e64 v0, s12, 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define <12 x float> @bitcast_v6f64_to_v12f32(<6 x double> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v12f32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB22_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; SI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; SI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; SI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; SI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; SI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; SI-NEXT: .LBB22_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v12f32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB22_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; VI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; VI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; VI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; VI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; VI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; VI-NEXT: .LBB22_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v12f32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB22_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX9-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX9-NEXT: .LBB22_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v12f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB22_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX11-NEXT: .LBB22_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define inreg <12 x float> @bitcast_v6f64_to_v12f32_scalar(<6 x double> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v12f32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB23_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB23_2: ; %cmp.true
; SI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; SI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; SI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; SI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; SI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; SI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v12f32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB23_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB23_2: ; %cmp.true
; VI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; VI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; VI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; VI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; VI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; VI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v12f32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB23_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB23_2: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; GFX9-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v12f32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB23_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB23_2: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[10:11], s[22:23], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], s[20:21], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], s[18:19], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], s[16:17], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], s[14:15], 1.0
; GFX11-NEXT: v_add_f64 v[0:1], s[12:13], 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define <6 x i64> @bitcast_v12f32_to_v6i64(<12 x float> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v6i64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB24_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_f32_e32 v11, 1.0, v11
; SI-NEXT: v_add_f32_e32 v10, 1.0, v10
; SI-NEXT: v_add_f32_e32 v9, 1.0, v9
; SI-NEXT: v_add_f32_e32 v8, 1.0, v8
; SI-NEXT: v_add_f32_e32 v7, 1.0, v7
; SI-NEXT: v_add_f32_e32 v6, 1.0, v6
; SI-NEXT: v_add_f32_e32 v5, 1.0, v5
; SI-NEXT: v_add_f32_e32 v4, 1.0, v4
; SI-NEXT: v_add_f32_e32 v3, 1.0, v3
; SI-NEXT: v_add_f32_e32 v2, 1.0, v2
; SI-NEXT: v_add_f32_e32 v1, 1.0, v1
; SI-NEXT: v_add_f32_e32 v0, 1.0, v0
; SI-NEXT: .LBB24_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v6i64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB24_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f32_e32 v11, 1.0, v11
; VI-NEXT: v_add_f32_e32 v10, 1.0, v10
; VI-NEXT: v_add_f32_e32 v9, 1.0, v9
; VI-NEXT: v_add_f32_e32 v8, 1.0, v8
; VI-NEXT: v_add_f32_e32 v7, 1.0, v7
; VI-NEXT: v_add_f32_e32 v6, 1.0, v6
; VI-NEXT: v_add_f32_e32 v5, 1.0, v5
; VI-NEXT: v_add_f32_e32 v4, 1.0, v4
; VI-NEXT: v_add_f32_e32 v3, 1.0, v3
; VI-NEXT: v_add_f32_e32 v2, 1.0, v2
; VI-NEXT: v_add_f32_e32 v1, 1.0, v1
; VI-NEXT: v_add_f32_e32 v0, 1.0, v0
; VI-NEXT: .LBB24_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v6i64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB24_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f32_e32 v11, 1.0, v11
; GFX9-NEXT: v_add_f32_e32 v10, 1.0, v10
; GFX9-NEXT: v_add_f32_e32 v9, 1.0, v9
; GFX9-NEXT: v_add_f32_e32 v8, 1.0, v8
; GFX9-NEXT: v_add_f32_e32 v7, 1.0, v7
; GFX9-NEXT: v_add_f32_e32 v6, 1.0, v6
; GFX9-NEXT: v_add_f32_e32 v5, 1.0, v5
; GFX9-NEXT: v_add_f32_e32 v4, 1.0, v4
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v3
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v2
; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
; GFX9-NEXT: v_add_f32_e32 v0, 1.0, v0
; GFX9-NEXT: .LBB24_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v6i64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_dual_add_f32 v11, 1.0, v11 :: v_dual_add_f32 v10, 1.0, v10
; GFX11-NEXT: v_dual_add_f32 v9, 1.0, v9 :: v_dual_add_f32 v8, 1.0, v8
; GFX11-NEXT: v_dual_add_f32 v7, 1.0, v7 :: v_dual_add_f32 v6, 1.0, v6
; GFX11-NEXT: v_dual_add_f32 v5, 1.0, v5 :: v_dual_add_f32 v4, 1.0, v4
; GFX11-NEXT: v_dual_add_f32 v3, 1.0, v3 :: v_dual_add_f32 v2, 1.0, v2
; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v0, 1.0, v0
; GFX11-NEXT: ; %bb.2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define inreg <6 x i64> @bitcast_v12f32_to_v6i64_scalar(<12 x float> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v6i64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB25_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: v_mov_b32_e32 v12, s28
; SI-NEXT: v_mov_b32_e32 v13, s29
; SI-NEXT: v_mov_b32_e32 v14, s30
; SI-NEXT: v_mov_b32_e32 v15, s31
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB25_2: ; %cmp.true
; SI-NEXT: v_add_f32_e64 v11, s27, 1.0
; SI-NEXT: v_add_f32_e64 v10, s26, 1.0
; SI-NEXT: v_add_f32_e64 v9, s25, 1.0
; SI-NEXT: v_add_f32_e64 v8, s24, 1.0
; SI-NEXT: v_add_f32_e64 v7, s23, 1.0
; SI-NEXT: v_add_f32_e64 v6, s22, 1.0
; SI-NEXT: v_add_f32_e64 v5, s21, 1.0
; SI-NEXT: v_add_f32_e64 v4, s20, 1.0
; SI-NEXT: v_add_f32_e64 v3, s19, 1.0
; SI-NEXT: v_add_f32_e64 v2, s18, 1.0
; SI-NEXT: v_add_f32_e64 v1, s17, 1.0
; SI-NEXT: v_add_f32_e64 v0, s16, 1.0
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v6i64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB25_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB25_2: ; %cmp.true
; VI-NEXT: v_add_f32_e64 v11, s27, 1.0
; VI-NEXT: v_add_f32_e64 v10, s26, 1.0
; VI-NEXT: v_add_f32_e64 v9, s25, 1.0
; VI-NEXT: v_add_f32_e64 v8, s24, 1.0
; VI-NEXT: v_add_f32_e64 v7, s23, 1.0
; VI-NEXT: v_add_f32_e64 v6, s22, 1.0
; VI-NEXT: v_add_f32_e64 v5, s21, 1.0
; VI-NEXT: v_add_f32_e64 v4, s20, 1.0
; VI-NEXT: v_add_f32_e64 v3, s19, 1.0
; VI-NEXT: v_add_f32_e64 v2, s18, 1.0
; VI-NEXT: v_add_f32_e64 v1, s17, 1.0
; VI-NEXT: v_add_f32_e64 v0, s16, 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v6i64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB25_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB25_2: ; %cmp.true
; GFX9-NEXT: v_add_f32_e64 v11, s27, 1.0
; GFX9-NEXT: v_add_f32_e64 v10, s26, 1.0
; GFX9-NEXT: v_add_f32_e64 v9, s25, 1.0
; GFX9-NEXT: v_add_f32_e64 v8, s24, 1.0
; GFX9-NEXT: v_add_f32_e64 v7, s23, 1.0
; GFX9-NEXT: v_add_f32_e64 v6, s22, 1.0
; GFX9-NEXT: v_add_f32_e64 v5, s21, 1.0
; GFX9-NEXT: v_add_f32_e64 v4, s20, 1.0
; GFX9-NEXT: v_add_f32_e64 v3, s19, 1.0
; GFX9-NEXT: v_add_f32_e64 v2, s18, 1.0
; GFX9-NEXT: v_add_f32_e64 v1, s17, 1.0
; GFX9-NEXT: v_add_f32_e64 v0, s16, 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v6i64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB25_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB25_2: ; %cmp.true
; GFX11-NEXT: v_add_f32_e64 v11, s23, 1.0
; GFX11-NEXT: v_add_f32_e64 v10, s22, 1.0
; GFX11-NEXT: v_add_f32_e64 v9, s21, 1.0
; GFX11-NEXT: v_add_f32_e64 v8, s20, 1.0
; GFX11-NEXT: v_add_f32_e64 v7, s19, 1.0
; GFX11-NEXT: v_add_f32_e64 v6, s18, 1.0
; GFX11-NEXT: v_add_f32_e64 v5, s17, 1.0
; GFX11-NEXT: v_add_f32_e64 v4, s16, 1.0
; GFX11-NEXT: v_add_f32_e64 v3, s15, 1.0
; GFX11-NEXT: v_add_f32_e64 v2, s14, 1.0
; GFX11-NEXT: v_add_f32_e64 v1, s13, 1.0
; GFX11-NEXT: v_add_f32_e64 v0, s12, 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define <12 x float> @bitcast_v6i64_to_v12f32(<6 x i64> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v12f32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB26_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; SI-NEXT: .LBB26_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v12f32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB26_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: .LBB26_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v12f32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB26_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_co_u32_e32 v10, vcc, 3, v10
; GFX9-NEXT: v_addc_co_u32_e32 v11, vcc, 0, v11, vcc
; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, 3, v8
; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, 0, v9, vcc
; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, 3, v6
; GFX9-NEXT: v_addc_co_u32_e32 v7, vcc, 0, v7, vcc
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, 3, v4
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, 3, v2
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, 3, v0
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: .LBB26_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v12f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB26_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_co_u32 v10, vcc_lo, v10, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v11, null, 0, v11, vcc_lo
; GFX11-NEXT: v_add_co_u32 v8, vcc_lo, v8, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v9, null, 0, v9, vcc_lo
; GFX11-NEXT: v_add_co_u32 v6, vcc_lo, v6, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v7, null, 0, v7, vcc_lo
; GFX11-NEXT: v_add_co_u32 v4, vcc_lo, v4, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v5, null, 0, v5, vcc_lo
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: .LBB26_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define inreg <12 x float> @bitcast_v6i64_to_v12f32_scalar(<6 x i64> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v12f32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc1 .LBB27_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: s_add_u32 s26, s26, 3
; SI-NEXT: s_addc_u32 s27, s27, 0
; SI-NEXT: s_add_u32 s24, s24, 3
; SI-NEXT: s_addc_u32 s25, s25, 0
; SI-NEXT: s_add_u32 s22, s22, 3
; SI-NEXT: s_addc_u32 s23, s23, 0
; SI-NEXT: s_add_u32 s20, s20, 3
; SI-NEXT: s_addc_u32 s21, s21, 0
; SI-NEXT: s_add_u32 s18, s18, 3
; SI-NEXT: s_addc_u32 s19, s19, 0
; SI-NEXT: s_add_u32 s16, s16, 3
; SI-NEXT: s_addc_u32 s17, s17, 0
; SI-NEXT: .LBB27_2: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v12f32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB27_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_u32 s26, s26, 3
; VI-NEXT: s_addc_u32 s27, s27, 0
; VI-NEXT: s_add_u32 s24, s24, 3
; VI-NEXT: s_addc_u32 s25, s25, 0
; VI-NEXT: s_add_u32 s22, s22, 3
; VI-NEXT: s_addc_u32 s23, s23, 0
; VI-NEXT: s_add_u32 s20, s20, 3
; VI-NEXT: s_addc_u32 s21, s21, 0
; VI-NEXT: s_add_u32 s18, s18, 3
; VI-NEXT: s_addc_u32 s19, s19, 0
; VI-NEXT: s_add_u32 s16, s16, 3
; VI-NEXT: s_addc_u32 s17, s17, 0
; VI-NEXT: .LBB27_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v12f32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB27_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_u32 s26, s26, 3
; GFX9-NEXT: s_addc_u32 s27, s27, 0
; GFX9-NEXT: s_add_u32 s24, s24, 3
; GFX9-NEXT: s_addc_u32 s25, s25, 0
; GFX9-NEXT: s_add_u32 s22, s22, 3
; GFX9-NEXT: s_addc_u32 s23, s23, 0
; GFX9-NEXT: s_add_u32 s20, s20, 3
; GFX9-NEXT: s_addc_u32 s21, s21, 0
; GFX9-NEXT: s_add_u32 s18, s18, 3
; GFX9-NEXT: s_addc_u32 s19, s19, 0
; GFX9-NEXT: s_add_u32 s16, s16, 3
; GFX9-NEXT: s_addc_u32 s17, s17, 0
; GFX9-NEXT: .LBB27_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v12f32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB27_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_u32 s22, s22, 3
; GFX11-NEXT: s_addc_u32 s23, s23, 0
; GFX11-NEXT: s_add_u32 s20, s20, 3
; GFX11-NEXT: s_addc_u32 s21, s21, 0
; GFX11-NEXT: s_add_u32 s18, s18, 3
; GFX11-NEXT: s_addc_u32 s19, s19, 0
; GFX11-NEXT: s_add_u32 s16, s16, 3
; GFX11-NEXT: s_addc_u32 s17, s17, 0
; GFX11-NEXT: s_add_u32 s2, s2, 3
; GFX11-NEXT: s_addc_u32 s3, s3, 0
; GFX11-NEXT: s_add_u32 s0, s0, 3
; GFX11-NEXT: s_addc_u32 s1, s1, 0
; GFX11-NEXT: .LBB27_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define <24 x i16> @bitcast_v12f32_to_v24i16(<12 x float> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v24i16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB28_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB28_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB28_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_f32_e32 v1, 1.0, v1
; SI-NEXT: v_add_f32_e32 v0, 1.0, v0
; SI-NEXT: v_add_f32_e32 v3, 1.0, v3
; SI-NEXT: v_add_f32_e32 v2, 1.0, v2
; SI-NEXT: v_add_f32_e32 v5, 1.0, v5
; SI-NEXT: v_add_f32_e32 v4, 1.0, v4
; SI-NEXT: v_add_f32_e32 v7, 1.0, v7
; SI-NEXT: v_add_f32_e32 v6, 1.0, v6
; SI-NEXT: v_add_f32_e32 v9, 1.0, v9
; SI-NEXT: v_add_f32_e32 v8, 1.0, v8
; SI-NEXT: v_add_f32_e32 v11, 1.0, v11
; SI-NEXT: v_add_f32_e32 v10, 1.0, v10
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB28_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v0, v0, v20
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v17
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v18
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v17
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v24i16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB28_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f32_e32 v11, 1.0, v11
; VI-NEXT: v_add_f32_e32 v10, 1.0, v10
; VI-NEXT: v_add_f32_e32 v9, 1.0, v9
; VI-NEXT: v_add_f32_e32 v8, 1.0, v8
; VI-NEXT: v_add_f32_e32 v7, 1.0, v7
; VI-NEXT: v_add_f32_e32 v6, 1.0, v6
; VI-NEXT: v_add_f32_e32 v5, 1.0, v5
; VI-NEXT: v_add_f32_e32 v4, 1.0, v4
; VI-NEXT: v_add_f32_e32 v3, 1.0, v3
; VI-NEXT: v_add_f32_e32 v2, 1.0, v2
; VI-NEXT: v_add_f32_e32 v1, 1.0, v1
; VI-NEXT: v_add_f32_e32 v0, 1.0, v0
; VI-NEXT: .LBB28_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v24i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB28_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f32_e32 v11, 1.0, v11
; GFX9-NEXT: v_add_f32_e32 v10, 1.0, v10
; GFX9-NEXT: v_add_f32_e32 v9, 1.0, v9
; GFX9-NEXT: v_add_f32_e32 v8, 1.0, v8
; GFX9-NEXT: v_add_f32_e32 v7, 1.0, v7
; GFX9-NEXT: v_add_f32_e32 v6, 1.0, v6
; GFX9-NEXT: v_add_f32_e32 v5, 1.0, v5
; GFX9-NEXT: v_add_f32_e32 v4, 1.0, v4
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v3
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v2
; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
; GFX9-NEXT: v_add_f32_e32 v0, 1.0, v0
; GFX9-NEXT: .LBB28_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v24i16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_dual_add_f32 v11, 1.0, v11 :: v_dual_add_f32 v10, 1.0, v10
; GFX11-NEXT: v_dual_add_f32 v9, 1.0, v9 :: v_dual_add_f32 v8, 1.0, v8
; GFX11-NEXT: v_dual_add_f32 v7, 1.0, v7 :: v_dual_add_f32 v6, 1.0, v6
; GFX11-NEXT: v_dual_add_f32 v5, 1.0, v5 :: v_dual_add_f32 v4, 1.0, v4
; GFX11-NEXT: v_dual_add_f32 v3, 1.0, v3 :: v_dual_add_f32 v2, 1.0, v2
; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v0, 1.0, v0
; GFX11-NEXT: ; %bb.2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define inreg <24 x i16> @bitcast_v12f32_to_v24i16_scalar(<12 x float> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v24i16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB29_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: v_mov_b32_e32 v24, s43
; SI-NEXT: v_mov_b32_e32 v23, s42
; SI-NEXT: v_mov_b32_e32 v22, s41
; SI-NEXT: v_mov_b32_e32 v21, s40
; SI-NEXT: v_mov_b32_e32 v20, s29
; SI-NEXT: v_mov_b32_e32 v19, s28
; SI-NEXT: v_mov_b32_e32 v17, s14
; SI-NEXT: v_mov_b32_e32 v16, s12
; SI-NEXT: v_mov_b32_e32 v15, s10
; SI-NEXT: v_mov_b32_e32 v14, s8
; SI-NEXT: v_mov_b32_e32 v13, s6
; SI-NEXT: v_mov_b32_e32 v12, s4
; SI-NEXT: s_branch .LBB29_3
; SI-NEXT: .LBB29_2: ; %cmp.true
; SI-NEXT: v_add_f32_e64 v11, s27, 1.0
; SI-NEXT: v_add_f32_e64 v10, s26, 1.0
; SI-NEXT: v_add_f32_e64 v9, s25, 1.0
; SI-NEXT: v_add_f32_e64 v8, s24, 1.0
; SI-NEXT: v_lshr_b64 v[12:13], v[10:11], 16
; SI-NEXT: v_add_f32_e64 v7, s23, 1.0
; SI-NEXT: v_add_f32_e64 v6, s22, 1.0
; SI-NEXT: v_lshr_b64 v[13:14], v[8:9], 16
; SI-NEXT: v_add_f32_e64 v5, s21, 1.0
; SI-NEXT: v_add_f32_e64 v4, s20, 1.0
; SI-NEXT: v_lshr_b64 v[14:15], v[6:7], 16
; SI-NEXT: v_add_f32_e64 v3, s19, 1.0
; SI-NEXT: v_add_f32_e64 v2, s18, 1.0
; SI-NEXT: v_lshr_b64 v[15:16], v[4:5], 16
; SI-NEXT: v_add_f32_e64 v1, s17, 1.0
; SI-NEXT: v_add_f32_e64 v0, s16, 1.0
; SI-NEXT: v_lshr_b64 v[16:17], v[2:3], 16
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v20, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v24, 16, v1
; SI-NEXT: v_lshr_b64 v[17:18], v[0:1], 16
; SI-NEXT: .LBB29_3: ; %end
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_or_b32_e32 v0, v0, v17
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v24
; SI-NEXT: v_or_b32_e32 v2, v2, v16
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v16, 16, v23
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v22
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v21
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v20
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v19
; SI-NEXT: v_or_b32_e32 v1, v1, v17
; SI-NEXT: v_or_b32_e32 v3, v3, v16
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v24i16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB29_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB29_2: ; %cmp.true
; VI-NEXT: v_add_f32_e64 v11, s27, 1.0
; VI-NEXT: v_add_f32_e64 v10, s26, 1.0
; VI-NEXT: v_add_f32_e64 v9, s25, 1.0
; VI-NEXT: v_add_f32_e64 v8, s24, 1.0
; VI-NEXT: v_add_f32_e64 v7, s23, 1.0
; VI-NEXT: v_add_f32_e64 v6, s22, 1.0
; VI-NEXT: v_add_f32_e64 v5, s21, 1.0
; VI-NEXT: v_add_f32_e64 v4, s20, 1.0
; VI-NEXT: v_add_f32_e64 v3, s19, 1.0
; VI-NEXT: v_add_f32_e64 v2, s18, 1.0
; VI-NEXT: v_add_f32_e64 v1, s17, 1.0
; VI-NEXT: v_add_f32_e64 v0, s16, 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v24i16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB29_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB29_2: ; %cmp.true
; GFX9-NEXT: v_add_f32_e64 v11, s27, 1.0
; GFX9-NEXT: v_add_f32_e64 v10, s26, 1.0
; GFX9-NEXT: v_add_f32_e64 v9, s25, 1.0
; GFX9-NEXT: v_add_f32_e64 v8, s24, 1.0
; GFX9-NEXT: v_add_f32_e64 v7, s23, 1.0
; GFX9-NEXT: v_add_f32_e64 v6, s22, 1.0
; GFX9-NEXT: v_add_f32_e64 v5, s21, 1.0
; GFX9-NEXT: v_add_f32_e64 v4, s20, 1.0
; GFX9-NEXT: v_add_f32_e64 v3, s19, 1.0
; GFX9-NEXT: v_add_f32_e64 v2, s18, 1.0
; GFX9-NEXT: v_add_f32_e64 v1, s17, 1.0
; GFX9-NEXT: v_add_f32_e64 v0, s16, 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v24i16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB29_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB29_2: ; %cmp.true
; GFX11-NEXT: v_add_f32_e64 v11, s23, 1.0
; GFX11-NEXT: v_add_f32_e64 v10, s22, 1.0
; GFX11-NEXT: v_add_f32_e64 v9, s21, 1.0
; GFX11-NEXT: v_add_f32_e64 v8, s20, 1.0
; GFX11-NEXT: v_add_f32_e64 v7, s19, 1.0
; GFX11-NEXT: v_add_f32_e64 v6, s18, 1.0
; GFX11-NEXT: v_add_f32_e64 v5, s17, 1.0
; GFX11-NEXT: v_add_f32_e64 v4, s16, 1.0
; GFX11-NEXT: v_add_f32_e64 v3, s15, 1.0
; GFX11-NEXT: v_add_f32_e64 v2, s14, 1.0
; GFX11-NEXT: v_add_f32_e64 v1, s13, 1.0
; GFX11-NEXT: v_add_f32_e64 v0, s12, 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define <12 x float> @bitcast_v24i16_to_v12f32(<24 x i16> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v12f32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_mov_b32_e32 v24, v11
; SI-NEXT: v_mov_b32_e32 v13, v10
; SI-NEXT: v_mov_b32_e32 v14, v9
; SI-NEXT: v_mov_b32_e32 v15, v8
; SI-NEXT: v_mov_b32_e32 v16, v7
; SI-NEXT: v_mov_b32_e32 v17, v6
; SI-NEXT: v_mov_b32_e32 v18, v5
; SI-NEXT: v_mov_b32_e32 v19, v4
; SI-NEXT: v_mov_b32_e32 v20, v3
; SI-NEXT: v_mov_b32_e32 v21, v2
; SI-NEXT: v_mov_b32_e32 v22, v1
; SI-NEXT: v_mov_b32_e32 v23, v0
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v24
; SI-NEXT: v_lshrrev_b32_e32 v1, 16, v13
; SI-NEXT: v_lshrrev_b32_e32 v2, 16, v14
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v15
; SI-NEXT: v_lshrrev_b32_e32 v4, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v5, 16, v17
; SI-NEXT: v_lshrrev_b32_e32 v6, 16, v18
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v8, 16, v20
; SI-NEXT: v_lshrrev_b32_e32 v9, 16, v21
; SI-NEXT: v_lshrrev_b32_e32 v10, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v11, 16, v23
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: v_lshlrev_b32_e32 v35, 16, v11
; SI-NEXT: v_lshlrev_b32_e32 v34, 16, v10
; SI-NEXT: v_lshlrev_b32_e32 v33, 16, v9
; SI-NEXT: v_lshlrev_b32_e32 v32, 16, v8
; SI-NEXT: v_lshlrev_b32_e32 v31, 16, v7
; SI-NEXT: v_lshlrev_b32_e32 v30, 16, v6
; SI-NEXT: v_lshlrev_b32_e32 v29, 16, v5
; SI-NEXT: v_lshlrev_b32_e32 v28, 16, v4
; SI-NEXT: v_lshlrev_b32_e32 v27, 16, v3
; SI-NEXT: v_lshlrev_b32_e32 v26, 16, v2
; SI-NEXT: v_lshlrev_b32_e32 v25, 16, v1
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v0
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB30_3
; SI-NEXT: ; %bb.1: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB30_4
; SI-NEXT: .LBB30_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB30_3: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v23
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v22
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v21
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v20
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v19
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v18
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v17
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v16
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v15
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v14
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v13
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v24
; SI-NEXT: v_or_b32_e32 v0, v0, v35
; SI-NEXT: v_or_b32_e32 v1, v1, v34
; SI-NEXT: v_or_b32_e32 v2, v2, v33
; SI-NEXT: v_or_b32_e32 v3, v3, v32
; SI-NEXT: v_or_b32_e32 v4, v4, v31
; SI-NEXT: v_or_b32_e32 v5, v5, v30
; SI-NEXT: v_or_b32_e32 v6, v6, v29
; SI-NEXT: v_or_b32_e32 v7, v7, v28
; SI-NEXT: v_or_b32_e32 v8, v8, v27
; SI-NEXT: v_or_b32_e32 v9, v9, v26
; SI-NEXT: v_or_b32_e32 v10, v10, v25
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB30_2
; SI-NEXT: .LBB30_4: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v23
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v22
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v21
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v20
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v19
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v18
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v17
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v16
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v15
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v14
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v13
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v24
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_or_b32_e32 v0, v35, v0
; SI-NEXT: v_or_b32_e32 v1, v34, v1
; SI-NEXT: v_or_b32_e32 v2, v33, v2
; SI-NEXT: v_or_b32_e32 v3, v32, v3
; SI-NEXT: v_or_b32_e32 v4, v31, v4
; SI-NEXT: v_or_b32_e32 v5, v30, v5
; SI-NEXT: v_or_b32_e32 v6, v29, v6
; SI-NEXT: v_or_b32_e32 v7, v28, v7
; SI-NEXT: v_or_b32_e32 v8, v27, v8
; SI-NEXT: v_or_b32_e32 v9, v26, v9
; SI-NEXT: v_or_b32_e32 v10, v25, v10
; SI-NEXT: v_or_b32_e32 v11, v12, v11
; SI-NEXT: v_add_i32_e32 v0, vcc, 0x30000, v0
; SI-NEXT: v_add_i32_e32 v1, vcc, 0x30000, v1
; SI-NEXT: v_add_i32_e32 v2, vcc, 0x30000, v2
; SI-NEXT: v_add_i32_e32 v3, vcc, 0x30000, v3
; SI-NEXT: v_add_i32_e32 v4, vcc, 0x30000, v4
; SI-NEXT: v_add_i32_e32 v5, vcc, 0x30000, v5
; SI-NEXT: v_add_i32_e32 v6, vcc, 0x30000, v6
; SI-NEXT: v_add_i32_e32 v7, vcc, 0x30000, v7
; SI-NEXT: v_add_i32_e32 v8, vcc, 0x30000, v8
; SI-NEXT: v_add_i32_e32 v9, vcc, 0x30000, v9
; SI-NEXT: v_add_i32_e32 v10, vcc, 0x30000, v10
; SI-NEXT: v_add_i32_e32 v11, vcc, 0x30000, v11
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v12f32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB30_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v13, 3
; VI-NEXT: v_add_u16_e32 v12, 3, v11
; VI-NEXT: v_add_u16_sdwa v11, v11, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v11, v12, v11
; VI-NEXT: v_add_u16_e32 v12, 3, v10
; VI-NEXT: v_add_u16_sdwa v10, v10, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v10, v12, v10
; VI-NEXT: v_add_u16_e32 v12, 3, v9
; VI-NEXT: v_add_u16_sdwa v9, v9, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v9, v12, v9
; VI-NEXT: v_add_u16_e32 v12, 3, v8
; VI-NEXT: v_add_u16_sdwa v8, v8, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v8, v12, v8
; VI-NEXT: v_add_u16_e32 v12, 3, v7
; VI-NEXT: v_add_u16_sdwa v7, v7, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v7, v12, v7
; VI-NEXT: v_add_u16_e32 v12, 3, v6
; VI-NEXT: v_add_u16_sdwa v6, v6, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v6, v12, v6
; VI-NEXT: v_add_u16_e32 v12, 3, v5
; VI-NEXT: v_add_u16_sdwa v5, v5, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v5, v12, v5
; VI-NEXT: v_add_u16_e32 v12, 3, v4
; VI-NEXT: v_add_u16_sdwa v4, v4, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v4, v12, v4
; VI-NEXT: v_add_u16_e32 v12, 3, v3
; VI-NEXT: v_add_u16_sdwa v3, v3, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v3, v12, v3
; VI-NEXT: v_add_u16_e32 v12, 3, v2
; VI-NEXT: v_add_u16_sdwa v2, v2, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v2, v12, v2
; VI-NEXT: v_add_u16_e32 v12, 3, v1
; VI-NEXT: v_add_u16_sdwa v1, v1, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v1, v12, v1
; VI-NEXT: v_add_u16_e32 v12, 3, v0
; VI-NEXT: v_add_u16_sdwa v0, v0, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v0, v12, v0
; VI-NEXT: .LBB30_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v12f32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB30_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB30_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v12f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB30_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB30_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define inreg <12 x float> @bitcast_v24i16_to_v12f32_scalar(<24 x i16> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v12f32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s12, s27, 16
; SI-NEXT: s_lshr_b32 s14, s26, 16
; SI-NEXT: s_lshr_b32 s15, s25, 16
; SI-NEXT: s_lshr_b32 s13, s24, 16
; SI-NEXT: s_lshr_b32 s11, s23, 16
; SI-NEXT: s_lshr_b32 s10, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s8, s20, 16
; SI-NEXT: s_lshr_b32 s7, s19, 16
; SI-NEXT: s_lshr_b32 s6, s18, 16
; SI-NEXT: s_lshr_b32 s5, s17, 16
; SI-NEXT: s_lshr_b32 s4, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB31_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s16, s4
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_or_b32 s5, s16, s5
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s16, s6
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_or_b32 s7, s16, s7
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s16, s8
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_or_b32 s9, s16, s9
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s16, s10
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_or_b32 s11, s16, s11
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_or_b32 s13, s16, s13
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_or_b32 s14, s16, s14
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s16, s12
; SI-NEXT: s_branch .LBB31_3
; SI-NEXT: .LBB31_2: ; %cmp.true
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_or_b32 s4, s4, s16
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_or_b32 s5, s5, s16
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_or_b32 s6, s6, s16
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_or_b32 s7, s7, s16
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_or_b32 s8, s8, s16
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_or_b32 s9, s9, s16
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_or_b32 s10, s10, s16
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_or_b32 s11, s11, s16
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_or_b32 s13, s13, s16
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_or_b32 s14, s14, s16
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s12, s16
; SI-NEXT: s_add_i32 s4, s4, 0x30000
; SI-NEXT: s_add_i32 s5, s5, 0x30000
; SI-NEXT: s_add_i32 s6, s6, 0x30000
; SI-NEXT: s_add_i32 s7, s7, 0x30000
; SI-NEXT: s_add_i32 s8, s8, 0x30000
; SI-NEXT: s_add_i32 s9, s9, 0x30000
; SI-NEXT: s_add_i32 s10, s10, 0x30000
; SI-NEXT: s_add_i32 s11, s11, 0x30000
; SI-NEXT: s_add_i32 s13, s13, 0x30000
; SI-NEXT: s_add_i32 s15, s15, 0x30000
; SI-NEXT: s_add_i32 s14, s14, 0x30000
; SI-NEXT: s_add_i32 s12, s12, 0x30000
; SI-NEXT: .LBB31_3: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: v_mov_b32_e32 v2, s6
; SI-NEXT: v_mov_b32_e32 v3, s7
; SI-NEXT: v_mov_b32_e32 v4, s8
; SI-NEXT: v_mov_b32_e32 v5, s9
; SI-NEXT: v_mov_b32_e32 v6, s10
; SI-NEXT: v_mov_b32_e32 v7, s11
; SI-NEXT: v_mov_b32_e32 v8, s13
; SI-NEXT: v_mov_b32_e32 v9, s15
; SI-NEXT: v_mov_b32_e32 v10, s14
; SI-NEXT: v_mov_b32_e32 v11, s12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v12f32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB31_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s5, s27, 3
; VI-NEXT: s_and_b32 s4, s27, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s26, 3
; VI-NEXT: s_add_i32 s27, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s26, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s25, 3
; VI-NEXT: s_add_i32 s26, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s25, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s24, 3
; VI-NEXT: s_add_i32 s25, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s24, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s23, 3
; VI-NEXT: s_add_i32 s24, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s23, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s22, 3
; VI-NEXT: s_add_i32 s23, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s22, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s21, 3
; VI-NEXT: s_add_i32 s22, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s21, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s20, 3
; VI-NEXT: s_add_i32 s21, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s20, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s19, 3
; VI-NEXT: s_add_i32 s20, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s19, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s18, 3
; VI-NEXT: s_add_i32 s19, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s18, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s17, 3
; VI-NEXT: s_add_i32 s18, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s17, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s16, 3
; VI-NEXT: s_add_i32 s17, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s16, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s16, s4, 0x30000
; VI-NEXT: .LBB31_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v12f32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB31_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB31_2: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, s27, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, s26, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, s25, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, s24, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, s23, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, s22, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, s21, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, s20, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, s19, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, s18, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, s17, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, s16, 3 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v12f32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB31_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB31_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, s23, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, s22, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, s21, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, s20, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, s19, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, s18, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, s17, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, s16, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, s15, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, s14, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, s13, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, s12, 3 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define <24 x half> @bitcast_v12f32_to_v24f16(<12 x float> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v24f16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB32_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB32_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB32_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_f32_e32 v1, 1.0, v1
; SI-NEXT: v_add_f32_e32 v0, 1.0, v0
; SI-NEXT: v_add_f32_e32 v3, 1.0, v3
; SI-NEXT: v_add_f32_e32 v2, 1.0, v2
; SI-NEXT: v_add_f32_e32 v5, 1.0, v5
; SI-NEXT: v_add_f32_e32 v4, 1.0, v4
; SI-NEXT: v_add_f32_e32 v7, 1.0, v7
; SI-NEXT: v_add_f32_e32 v6, 1.0, v6
; SI-NEXT: v_add_f32_e32 v9, 1.0, v9
; SI-NEXT: v_add_f32_e32 v8, 1.0, v8
; SI-NEXT: v_add_f32_e32 v11, 1.0, v11
; SI-NEXT: v_add_f32_e32 v10, 1.0, v10
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB32_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v0, v0, v20
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v17
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v18
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v17
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v24f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB32_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f32_e32 v11, 1.0, v11
; VI-NEXT: v_add_f32_e32 v10, 1.0, v10
; VI-NEXT: v_add_f32_e32 v9, 1.0, v9
; VI-NEXT: v_add_f32_e32 v8, 1.0, v8
; VI-NEXT: v_add_f32_e32 v7, 1.0, v7
; VI-NEXT: v_add_f32_e32 v6, 1.0, v6
; VI-NEXT: v_add_f32_e32 v5, 1.0, v5
; VI-NEXT: v_add_f32_e32 v4, 1.0, v4
; VI-NEXT: v_add_f32_e32 v3, 1.0, v3
; VI-NEXT: v_add_f32_e32 v2, 1.0, v2
; VI-NEXT: v_add_f32_e32 v1, 1.0, v1
; VI-NEXT: v_add_f32_e32 v0, 1.0, v0
; VI-NEXT: .LBB32_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v24f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB32_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f32_e32 v11, 1.0, v11
; GFX9-NEXT: v_add_f32_e32 v10, 1.0, v10
; GFX9-NEXT: v_add_f32_e32 v9, 1.0, v9
; GFX9-NEXT: v_add_f32_e32 v8, 1.0, v8
; GFX9-NEXT: v_add_f32_e32 v7, 1.0, v7
; GFX9-NEXT: v_add_f32_e32 v6, 1.0, v6
; GFX9-NEXT: v_add_f32_e32 v5, 1.0, v5
; GFX9-NEXT: v_add_f32_e32 v4, 1.0, v4
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v3
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v2
; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
; GFX9-NEXT: v_add_f32_e32 v0, 1.0, v0
; GFX9-NEXT: .LBB32_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v24f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_dual_add_f32 v11, 1.0, v11 :: v_dual_add_f32 v10, 1.0, v10
; GFX11-NEXT: v_dual_add_f32 v9, 1.0, v9 :: v_dual_add_f32 v8, 1.0, v8
; GFX11-NEXT: v_dual_add_f32 v7, 1.0, v7 :: v_dual_add_f32 v6, 1.0, v6
; GFX11-NEXT: v_dual_add_f32 v5, 1.0, v5 :: v_dual_add_f32 v4, 1.0, v4
; GFX11-NEXT: v_dual_add_f32 v3, 1.0, v3 :: v_dual_add_f32 v2, 1.0, v2
; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v0, 1.0, v0
; GFX11-NEXT: ; %bb.2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define inreg <24 x half> @bitcast_v12f32_to_v24f16_scalar(<12 x float> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v12f32_to_v24f16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB33_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: v_mov_b32_e32 v24, s43
; SI-NEXT: v_mov_b32_e32 v23, s42
; SI-NEXT: v_mov_b32_e32 v22, s41
; SI-NEXT: v_mov_b32_e32 v21, s40
; SI-NEXT: v_mov_b32_e32 v20, s29
; SI-NEXT: v_mov_b32_e32 v19, s28
; SI-NEXT: v_mov_b32_e32 v17, s14
; SI-NEXT: v_mov_b32_e32 v16, s12
; SI-NEXT: v_mov_b32_e32 v15, s10
; SI-NEXT: v_mov_b32_e32 v14, s8
; SI-NEXT: v_mov_b32_e32 v13, s6
; SI-NEXT: v_mov_b32_e32 v12, s4
; SI-NEXT: s_branch .LBB33_3
; SI-NEXT: .LBB33_2: ; %cmp.true
; SI-NEXT: v_add_f32_e64 v11, s27, 1.0
; SI-NEXT: v_add_f32_e64 v10, s26, 1.0
; SI-NEXT: v_add_f32_e64 v9, s25, 1.0
; SI-NEXT: v_add_f32_e64 v8, s24, 1.0
; SI-NEXT: v_lshr_b64 v[12:13], v[10:11], 16
; SI-NEXT: v_add_f32_e64 v7, s23, 1.0
; SI-NEXT: v_add_f32_e64 v6, s22, 1.0
; SI-NEXT: v_lshr_b64 v[13:14], v[8:9], 16
; SI-NEXT: v_add_f32_e64 v5, s21, 1.0
; SI-NEXT: v_add_f32_e64 v4, s20, 1.0
; SI-NEXT: v_lshr_b64 v[14:15], v[6:7], 16
; SI-NEXT: v_add_f32_e64 v3, s19, 1.0
; SI-NEXT: v_add_f32_e64 v2, s18, 1.0
; SI-NEXT: v_lshr_b64 v[15:16], v[4:5], 16
; SI-NEXT: v_add_f32_e64 v1, s17, 1.0
; SI-NEXT: v_add_f32_e64 v0, s16, 1.0
; SI-NEXT: v_lshr_b64 v[16:17], v[2:3], 16
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v20, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v24, 16, v1
; SI-NEXT: v_lshr_b64 v[17:18], v[0:1], 16
; SI-NEXT: .LBB33_3: ; %end
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_or_b32_e32 v0, v0, v17
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v24
; SI-NEXT: v_or_b32_e32 v2, v2, v16
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v16, 16, v23
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v22
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v21
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v20
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v19
; SI-NEXT: v_or_b32_e32 v1, v1, v17
; SI-NEXT: v_or_b32_e32 v3, v3, v16
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v12f32_to_v24f16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB33_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB33_2: ; %cmp.true
; VI-NEXT: v_add_f32_e64 v11, s27, 1.0
; VI-NEXT: v_add_f32_e64 v10, s26, 1.0
; VI-NEXT: v_add_f32_e64 v9, s25, 1.0
; VI-NEXT: v_add_f32_e64 v8, s24, 1.0
; VI-NEXT: v_add_f32_e64 v7, s23, 1.0
; VI-NEXT: v_add_f32_e64 v6, s22, 1.0
; VI-NEXT: v_add_f32_e64 v5, s21, 1.0
; VI-NEXT: v_add_f32_e64 v4, s20, 1.0
; VI-NEXT: v_add_f32_e64 v3, s19, 1.0
; VI-NEXT: v_add_f32_e64 v2, s18, 1.0
; VI-NEXT: v_add_f32_e64 v1, s17, 1.0
; VI-NEXT: v_add_f32_e64 v0, s16, 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v12f32_to_v24f16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB33_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB33_2: ; %cmp.true
; GFX9-NEXT: v_add_f32_e64 v11, s27, 1.0
; GFX9-NEXT: v_add_f32_e64 v10, s26, 1.0
; GFX9-NEXT: v_add_f32_e64 v9, s25, 1.0
; GFX9-NEXT: v_add_f32_e64 v8, s24, 1.0
; GFX9-NEXT: v_add_f32_e64 v7, s23, 1.0
; GFX9-NEXT: v_add_f32_e64 v6, s22, 1.0
; GFX9-NEXT: v_add_f32_e64 v5, s21, 1.0
; GFX9-NEXT: v_add_f32_e64 v4, s20, 1.0
; GFX9-NEXT: v_add_f32_e64 v3, s19, 1.0
; GFX9-NEXT: v_add_f32_e64 v2, s18, 1.0
; GFX9-NEXT: v_add_f32_e64 v1, s17, 1.0
; GFX9-NEXT: v_add_f32_e64 v0, s16, 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v12f32_to_v24f16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB33_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB33_2: ; %cmp.true
; GFX11-NEXT: v_add_f32_e64 v11, s23, 1.0
; GFX11-NEXT: v_add_f32_e64 v10, s22, 1.0
; GFX11-NEXT: v_add_f32_e64 v9, s21, 1.0
; GFX11-NEXT: v_add_f32_e64 v8, s20, 1.0
; GFX11-NEXT: v_add_f32_e64 v7, s19, 1.0
; GFX11-NEXT: v_add_f32_e64 v6, s18, 1.0
; GFX11-NEXT: v_add_f32_e64 v5, s17, 1.0
; GFX11-NEXT: v_add_f32_e64 v4, s16, 1.0
; GFX11-NEXT: v_add_f32_e64 v3, s15, 1.0
; GFX11-NEXT: v_add_f32_e64 v2, s14, 1.0
; GFX11-NEXT: v_add_f32_e64 v1, s13, 1.0
; GFX11-NEXT: v_add_f32_e64 v0, s12, 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <12 x float> %a, splat (float 1.000000e+00)
%a2 = bitcast <12 x float> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <12 x float> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define <12 x float> @bitcast_v24f16_to_v12f32(<24 x half> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v12f32:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_mov_b32_e32 v24, v11
; SI-NEXT: v_mov_b32_e32 v13, v10
; SI-NEXT: v_mov_b32_e32 v14, v9
; SI-NEXT: v_mov_b32_e32 v15, v8
; SI-NEXT: v_mov_b32_e32 v16, v7
; SI-NEXT: v_mov_b32_e32 v17, v6
; SI-NEXT: v_mov_b32_e32 v18, v5
; SI-NEXT: v_mov_b32_e32 v19, v4
; SI-NEXT: v_mov_b32_e32 v20, v3
; SI-NEXT: v_mov_b32_e32 v21, v2
; SI-NEXT: v_mov_b32_e32 v22, v1
; SI-NEXT: v_mov_b32_e32 v23, v0
; SI-NEXT: v_lshrrev_b32_e32 v25, 16, v24
; SI-NEXT: v_lshrrev_b32_e32 v26, 16, v13
; SI-NEXT: v_lshrrev_b32_e32 v27, 16, v14
; SI-NEXT: v_lshrrev_b32_e32 v28, 16, v15
; SI-NEXT: v_lshrrev_b32_e32 v29, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v30, 16, v17
; SI-NEXT: v_lshrrev_b32_e32 v31, 16, v18
; SI-NEXT: v_lshrrev_b32_e32 v32, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v33, 16, v20
; SI-NEXT: v_lshrrev_b32_e32 v34, 16, v21
; SI-NEXT: v_lshrrev_b32_e32 v35, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v36, 16, v23
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB34_3
; SI-NEXT: ; %bb.1: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB34_4
; SI-NEXT: .LBB34_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB34_3: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v23
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v36
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v22
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v35
; SI-NEXT: v_or_b32_e32 v0, v0, v1
; SI-NEXT: v_or_b32_e32 v1, v2, v3
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v21
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v34
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v20
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v33
; SI-NEXT: v_or_b32_e32 v3, v3, v4
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v19
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v32
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v18
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v31
; SI-NEXT: v_or_b32_e32 v5, v5, v6
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v17
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v30
; SI-NEXT: v_or_b32_e32 v6, v6, v7
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v16
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v29
; SI-NEXT: v_or_b32_e32 v7, v7, v8
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v15
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v28
; SI-NEXT: v_or_b32_e32 v8, v8, v9
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v14
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v27
; SI-NEXT: v_or_b32_e32 v9, v9, v10
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v13
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v26
; SI-NEXT: v_or_b32_e32 v10, v10, v11
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v24
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v25
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr36
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB34_2
; SI-NEXT: .LBB34_4: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v0, v36
; SI-NEXT: v_cvt_f32_f16_e32 v2, v35
; SI-NEXT: v_cvt_f32_f16_e32 v1, v23
; SI-NEXT: v_cvt_f32_f16_e32 v3, v22
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT: v_or_b32_e32 v0, v1, v0
; SI-NEXT: v_or_b32_e32 v1, v3, v2
; SI-NEXT: v_cvt_f32_f16_e32 v2, v34
; SI-NEXT: v_cvt_f32_f16_e32 v3, v21
; SI-NEXT: v_cvt_f32_f16_e32 v4, v33
; SI-NEXT: v_cvt_f32_f16_e32 v5, v20
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v6, v32
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT: v_or_b32_e32 v2, v3, v2
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_or_b32_e32 v3, v5, v3
; SI-NEXT: v_cvt_f32_f16_e32 v4, v19
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, v31
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, v18
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, v30
; SI-NEXT: v_cvt_f32_f16_e32 v8, v17
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_or_b32_e32 v5, v7, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, v29
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT: v_or_b32_e32 v6, v8, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, v16
; SI-NEXT: v_cvt_f32_f16_e32 v9, v28
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT: v_cvt_f32_f16_e32 v10, v15
; SI-NEXT: v_or_b32_e32 v7, v8, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v9
; SI-NEXT: v_cvt_f32_f16_e32 v9, v27
; SI-NEXT: v_cvt_f32_f16_e32 v11, v14
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_or_b32_e32 v8, v10, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, v26
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
; SI-NEXT: v_or_b32_e32 v9, v11, v9
; SI-NEXT: v_cvt_f32_f16_e32 v11, v13
; SI-NEXT: v_cvt_f32_f16_e32 v12, v25
; SI-NEXT: v_cvt_f32_f16_e32 v13, v24
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT: v_or_b32_e32 v10, v11, v10
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT: v_or_b32_e32 v11, v13, v11
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v12f32:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB34_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v12, 0x200
; VI-NEXT: v_add_f16_sdwa v13, v11, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v11, 0x200, v11
; VI-NEXT: v_or_b32_e32 v11, v11, v13
; VI-NEXT: v_add_f16_sdwa v13, v10, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v10, 0x200, v10
; VI-NEXT: v_or_b32_e32 v10, v10, v13
; VI-NEXT: v_add_f16_sdwa v13, v9, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v9, 0x200, v9
; VI-NEXT: v_or_b32_e32 v9, v9, v13
; VI-NEXT: v_add_f16_sdwa v13, v8, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v8, 0x200, v8
; VI-NEXT: v_or_b32_e32 v8, v8, v13
; VI-NEXT: v_add_f16_sdwa v13, v7, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v7, 0x200, v7
; VI-NEXT: v_or_b32_e32 v7, v7, v13
; VI-NEXT: v_add_f16_sdwa v13, v6, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v6, 0x200, v6
; VI-NEXT: v_or_b32_e32 v6, v6, v13
; VI-NEXT: v_add_f16_sdwa v13, v5, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v5, 0x200, v5
; VI-NEXT: v_or_b32_e32 v5, v5, v13
; VI-NEXT: v_add_f16_sdwa v13, v4, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v4, 0x200, v4
; VI-NEXT: v_or_b32_e32 v4, v4, v13
; VI-NEXT: v_add_f16_sdwa v13, v3, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v3, 0x200, v3
; VI-NEXT: v_or_b32_e32 v3, v3, v13
; VI-NEXT: v_add_f16_sdwa v13, v2, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, 0x200, v2
; VI-NEXT: v_or_b32_e32 v2, v2, v13
; VI-NEXT: v_add_f16_sdwa v13, v1, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v1, 0x200, v1
; VI-NEXT: v_add_f16_sdwa v12, v0, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, 0x200, v0
; VI-NEXT: v_or_b32_e32 v1, v1, v13
; VI-NEXT: v_or_b32_e32 v0, v0, v12
; VI-NEXT: .LBB34_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v12f32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB34_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_movk_i32 s6, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, v11, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, v10, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, v9, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, v8, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, v7, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, v6, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, v5, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, v4, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, v3, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, v2, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, v1, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, v0, s6 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB34_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v12f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB34_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, v11, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, v10, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, v9, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, v8, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, v7, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, v6, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, v5, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, v4, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, v3, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, v2, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, v1, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, v0, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB34_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define inreg <12 x float> @bitcast_v24f16_to_v12f32_scalar(<24 x half> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v12f32_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s15, s27, 16
; SI-NEXT: s_lshr_b32 s14, s26, 16
; SI-NEXT: s_lshr_b32 s13, s25, 16
; SI-NEXT: s_lshr_b32 s12, s24, 16
; SI-NEXT: s_lshr_b32 s11, s23, 16
; SI-NEXT: s_lshr_b32 s10, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s8, s20, 16
; SI-NEXT: s_lshr_b32 s7, s19, 16
; SI-NEXT: s_lshr_b32 s6, s18, 16
; SI-NEXT: s_lshr_b32 s5, s17, 16
; SI-NEXT: s_lshr_b32 s4, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB35_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s16, s4
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_or_b32 s5, s16, s5
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s16, s6
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_or_b32 s7, s16, s7
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s16, s8
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_or_b32 s9, s16, s9
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s16, s10
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_or_b32 s11, s16, s11
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s16, s12
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_or_b32 s13, s16, s13
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_or_b32 s14, s16, s14
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: v_mov_b32_e32 v2, s6
; SI-NEXT: v_mov_b32_e32 v3, s7
; SI-NEXT: v_mov_b32_e32 v4, s8
; SI-NEXT: v_mov_b32_e32 v5, s9
; SI-NEXT: v_mov_b32_e32 v6, s10
; SI-NEXT: v_mov_b32_e32 v7, s11
; SI-NEXT: v_mov_b32_e32 v8, s12
; SI-NEXT: v_mov_b32_e32 v9, s13
; SI-NEXT: v_mov_b32_e32 v10, s14
; SI-NEXT: v_mov_b32_e32 v11, s15
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB35_2: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v0, s4
; SI-NEXT: v_cvt_f32_f16_e32 v1, s16
; SI-NEXT: v_cvt_f32_f16_e32 v2, s5
; SI-NEXT: v_cvt_f32_f16_e32 v3, s17
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_or_b32_e32 v0, v1, v0
; SI-NEXT: v_cvt_f32_f16_e32 v1, s6
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f32_f16_e32 v5, s19
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v4, v1
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
; SI-NEXT: v_cvt_f32_f16_e32 v2, s18
; SI-NEXT: v_or_b32_e32 v1, v3, v1
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_cvt_f32_f16_e32 v4, s7
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v6, s8
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_or_b32_e32 v3, v5, v3
; SI-NEXT: v_cvt_f32_f16_e32 v4, s20
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, s9
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, s21
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, s10
; SI-NEXT: v_cvt_f32_f16_e32 v8, s22
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_or_b32_e32 v5, v7, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, s11
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT: v_or_b32_e32 v6, v8, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, s23
; SI-NEXT: v_cvt_f32_f16_e32 v9, s12
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT: v_cvt_f32_f16_e32 v10, s24
; SI-NEXT: v_or_b32_e32 v7, v8, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v9
; SI-NEXT: v_cvt_f32_f16_e32 v9, s13
; SI-NEXT: v_cvt_f32_f16_e32 v11, s25
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_or_b32_e32 v8, v10, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, s14
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
; SI-NEXT: v_or_b32_e32 v9, v11, v9
; SI-NEXT: v_cvt_f32_f16_e32 v11, s26
; SI-NEXT: v_cvt_f32_f16_e32 v12, s15
; SI-NEXT: v_cvt_f32_f16_e32 v13, s27
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT: v_or_b32_e32 v10, v11, v10
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT: v_or_b32_e32 v11, v13, v11
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v12f32_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB35_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB35_2: ; %cmp.true
; VI-NEXT: s_lshr_b32 s4, s27, 16
; VI-NEXT: v_mov_b32_e32 v0, 0x200
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s27, v0
; VI-NEXT: s_lshr_b32 s4, s26, 16
; VI-NEXT: v_or_b32_e32 v11, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s26, v0
; VI-NEXT: s_lshr_b32 s4, s25, 16
; VI-NEXT: v_or_b32_e32 v10, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s25, v0
; VI-NEXT: s_lshr_b32 s4, s24, 16
; VI-NEXT: v_or_b32_e32 v9, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s24, v0
; VI-NEXT: s_lshr_b32 s4, s23, 16
; VI-NEXT: v_or_b32_e32 v8, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s23, v0
; VI-NEXT: s_lshr_b32 s4, s22, 16
; VI-NEXT: v_or_b32_e32 v7, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s22, v0
; VI-NEXT: s_lshr_b32 s4, s21, 16
; VI-NEXT: v_or_b32_e32 v6, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s21, v0
; VI-NEXT: s_lshr_b32 s4, s20, 16
; VI-NEXT: v_or_b32_e32 v5, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s20, v0
; VI-NEXT: s_lshr_b32 s4, s19, 16
; VI-NEXT: v_or_b32_e32 v4, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s19, v0
; VI-NEXT: s_lshr_b32 s4, s18, 16
; VI-NEXT: v_or_b32_e32 v3, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s18, v0
; VI-NEXT: s_lshr_b32 s4, s17, 16
; VI-NEXT: v_or_b32_e32 v2, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v12, s17, v0
; VI-NEXT: s_lshr_b32 s4, s16, 16
; VI-NEXT: v_or_b32_e32 v1, v12, v1
; VI-NEXT: v_mov_b32_e32 v12, s4
; VI-NEXT: v_add_f16_sdwa v12, v12, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, s16, v0
; VI-NEXT: v_or_b32_e32 v0, v0, v12
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v12f32_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB35_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB35_2: ; %cmp.true
; GFX9-NEXT: v_mov_b32_e32 v0, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, s27, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, s26, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, s25, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, s24, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, s23, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, s22, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, s21, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, s20, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, s19, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, s18, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, s17, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, s16, v0 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v12f32_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB35_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB35_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, s23, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, s22, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, s21, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, s20, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, s19, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, s18, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, s17, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, s16, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, s15, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, s14, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, s13, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, s12, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <12 x float>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <12 x float>
br label %end
end:
%phi = phi <12 x float> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <12 x float> %phi
}
define <6 x i64> @bitcast_v6f64_to_v6i64(<6 x double> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v6i64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB36_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; SI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; SI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; SI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; SI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; SI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; SI-NEXT: .LBB36_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v6i64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB36_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; VI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; VI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; VI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; VI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; VI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; VI-NEXT: .LBB36_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v6i64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB36_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX9-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX9-NEXT: .LBB36_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v6i64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB36_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX11-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX11-NEXT: .LBB36_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define inreg <6 x i64> @bitcast_v6f64_to_v6i64_scalar(<6 x double> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v6i64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB37_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: v_mov_b32_e32 v12, s28
; SI-NEXT: v_mov_b32_e32 v13, s29
; SI-NEXT: v_mov_b32_e32 v14, s30
; SI-NEXT: v_mov_b32_e32 v15, s31
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB37_2: ; %cmp.true
; SI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; SI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; SI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; SI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; SI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; SI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v6i64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB37_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB37_2: ; %cmp.true
; VI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; VI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; VI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; VI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; VI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; VI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v6i64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB37_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB37_2: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; GFX9-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v6i64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB37_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB37_2: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[0:1], s[12:13], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], s[14:15], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], s[16:17], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], s[18:19], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], s[20:21], 1.0
; GFX11-NEXT: v_add_f64 v[10:11], s[22:23], 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define <6 x double> @bitcast_v6i64_to_v6f64(<6 x i64> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v6f64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB38_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; SI-NEXT: .LBB38_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v6f64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB38_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; VI-NEXT: .LBB38_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v6f64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB38_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, 3, v0
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, 3, v2
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, 3, v4
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, 3, v6
; GFX9-NEXT: v_addc_co_u32_e32 v7, vcc, 0, v7, vcc
; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, 3, v8
; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, 0, v9, vcc
; GFX9-NEXT: v_add_co_u32_e32 v10, vcc, 3, v10
; GFX9-NEXT: v_addc_co_u32_e32 v11, vcc, 0, v11, vcc
; GFX9-NEXT: .LBB38_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v6f64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB38_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: v_add_co_u32 v4, vcc_lo, v4, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v5, null, 0, v5, vcc_lo
; GFX11-NEXT: v_add_co_u32 v6, vcc_lo, v6, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v7, null, 0, v7, vcc_lo
; GFX11-NEXT: v_add_co_u32 v8, vcc_lo, v8, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v9, null, 0, v9, vcc_lo
; GFX11-NEXT: v_add_co_u32 v10, vcc_lo, v10, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v11, null, 0, v11, vcc_lo
; GFX11-NEXT: .LBB38_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define inreg <6 x double> @bitcast_v6i64_to_v6f64_scalar(<6 x i64> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v6f64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc1 .LBB39_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: s_add_u32 s16, s16, 3
; SI-NEXT: s_addc_u32 s17, s17, 0
; SI-NEXT: s_add_u32 s18, s18, 3
; SI-NEXT: s_addc_u32 s19, s19, 0
; SI-NEXT: s_add_u32 s20, s20, 3
; SI-NEXT: s_addc_u32 s21, s21, 0
; SI-NEXT: s_add_u32 s22, s22, 3
; SI-NEXT: s_addc_u32 s23, s23, 0
; SI-NEXT: s_add_u32 s24, s24, 3
; SI-NEXT: s_addc_u32 s25, s25, 0
; SI-NEXT: s_add_u32 s26, s26, 3
; SI-NEXT: s_addc_u32 s27, s27, 0
; SI-NEXT: .LBB39_2: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v6f64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB39_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_u32 s16, s16, 3
; VI-NEXT: s_addc_u32 s17, s17, 0
; VI-NEXT: s_add_u32 s18, s18, 3
; VI-NEXT: s_addc_u32 s19, s19, 0
; VI-NEXT: s_add_u32 s20, s20, 3
; VI-NEXT: s_addc_u32 s21, s21, 0
; VI-NEXT: s_add_u32 s22, s22, 3
; VI-NEXT: s_addc_u32 s23, s23, 0
; VI-NEXT: s_add_u32 s24, s24, 3
; VI-NEXT: s_addc_u32 s25, s25, 0
; VI-NEXT: s_add_u32 s26, s26, 3
; VI-NEXT: s_addc_u32 s27, s27, 0
; VI-NEXT: .LBB39_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v6f64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB39_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_u32 s16, s16, 3
; GFX9-NEXT: s_addc_u32 s17, s17, 0
; GFX9-NEXT: s_add_u32 s18, s18, 3
; GFX9-NEXT: s_addc_u32 s19, s19, 0
; GFX9-NEXT: s_add_u32 s20, s20, 3
; GFX9-NEXT: s_addc_u32 s21, s21, 0
; GFX9-NEXT: s_add_u32 s22, s22, 3
; GFX9-NEXT: s_addc_u32 s23, s23, 0
; GFX9-NEXT: s_add_u32 s24, s24, 3
; GFX9-NEXT: s_addc_u32 s25, s25, 0
; GFX9-NEXT: s_add_u32 s26, s26, 3
; GFX9-NEXT: s_addc_u32 s27, s27, 0
; GFX9-NEXT: .LBB39_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v6f64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB39_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_u32 s0, s0, 3
; GFX11-NEXT: s_addc_u32 s1, s1, 0
; GFX11-NEXT: s_add_u32 s2, s2, 3
; GFX11-NEXT: s_addc_u32 s3, s3, 0
; GFX11-NEXT: s_add_u32 s16, s16, 3
; GFX11-NEXT: s_addc_u32 s17, s17, 0
; GFX11-NEXT: s_add_u32 s18, s18, 3
; GFX11-NEXT: s_addc_u32 s19, s19, 0
; GFX11-NEXT: s_add_u32 s20, s20, 3
; GFX11-NEXT: s_addc_u32 s21, s21, 0
; GFX11-NEXT: s_add_u32 s22, s22, 3
; GFX11-NEXT: s_addc_u32 s23, s23, 0
; GFX11-NEXT: .LBB39_2: ; %end
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define <24 x i16> @bitcast_v6f64_to_v24i16(<6 x double> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v24i16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB40_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v18, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v17, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB40_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB40_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; SI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; SI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; SI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; SI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; SI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v18, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v17, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB40_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v18, 16, v18
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_or_b32_e32 v0, v0, v20
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v18
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v18, 16, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v17
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v18
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v24i16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB40_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; VI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; VI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; VI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; VI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; VI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; VI-NEXT: .LBB40_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v24i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB40_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX9-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX9-NEXT: .LBB40_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v24i16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB40_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX11-NEXT: .LBB40_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define inreg <24 x i16> @bitcast_v6f64_to_v24i16_scalar(<6 x double> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v24i16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB41_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v19, s28
; SI-NEXT: v_mov_b32_e32 v20, s29
; SI-NEXT: v_mov_b32_e32 v21, s40
; SI-NEXT: v_mov_b32_e32 v22, s41
; SI-NEXT: v_mov_b32_e32 v23, s42
; SI-NEXT: v_mov_b32_e32 v24, s43
; SI-NEXT: v_mov_b32_e32 v17, s14
; SI-NEXT: v_mov_b32_e32 v16, s12
; SI-NEXT: v_mov_b32_e32 v15, s10
; SI-NEXT: v_mov_b32_e32 v14, s8
; SI-NEXT: v_mov_b32_e32 v13, s6
; SI-NEXT: v_mov_b32_e32 v12, s4
; SI-NEXT: s_branch .LBB41_3
; SI-NEXT: .LBB41_2: ; %cmp.true
; SI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; SI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; SI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; SI-NEXT: v_lshr_b64 v[12:13], v[10:11], 16
; SI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; SI-NEXT: v_lshr_b64 v[13:14], v[8:9], 16
; SI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; SI-NEXT: v_lshr_b64 v[14:15], v[6:7], 16
; SI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; SI-NEXT: v_lshr_b64 v[15:16], v[4:5], 16
; SI-NEXT: v_lshr_b64 v[16:17], v[2:3], 16
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v20, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v24, 16, v1
; SI-NEXT: v_lshr_b64 v[17:18], v[0:1], 16
; SI-NEXT: .LBB41_3: ; %end
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v0, v0, v17
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v24
; SI-NEXT: v_or_b32_e32 v2, v2, v16
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v16, 16, v23
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v22
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v21
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v20
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v19
; SI-NEXT: v_or_b32_e32 v1, v1, v17
; SI-NEXT: v_or_b32_e32 v3, v3, v16
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v24i16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB41_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB41_2: ; %cmp.true
; VI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; VI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; VI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; VI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; VI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; VI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v24i16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB41_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB41_2: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; GFX9-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v24i16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB41_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB41_2: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[10:11], s[22:23], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], s[20:21], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], s[18:19], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], s[16:17], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], s[14:15], 1.0
; GFX11-NEXT: v_add_f64 v[0:1], s[12:13], 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define <6 x double> @bitcast_v24i16_to_v6f64(<24 x i16> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v6f64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_mov_b32_e32 v16, v11
; SI-NEXT: v_mov_b32_e32 v17, v10
; SI-NEXT: v_mov_b32_e32 v18, v9
; SI-NEXT: v_mov_b32_e32 v19, v8
; SI-NEXT: v_mov_b32_e32 v20, v7
; SI-NEXT: v_mov_b32_e32 v21, v6
; SI-NEXT: v_mov_b32_e32 v22, v5
; SI-NEXT: v_mov_b32_e32 v23, v4
; SI-NEXT: v_mov_b32_e32 v24, v3
; SI-NEXT: v_mov_b32_e32 v25, v2
; SI-NEXT: v_mov_b32_e32 v26, v1
; SI-NEXT: v_mov_b32_e32 v27, v0
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v1, 16, v17
; SI-NEXT: v_lshrrev_b32_e32 v2, 16, v18
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v4, 16, v20
; SI-NEXT: v_lshrrev_b32_e32 v5, 16, v21
; SI-NEXT: v_lshrrev_b32_e32 v6, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v23
; SI-NEXT: v_lshrrev_b32_e32 v8, 16, v24
; SI-NEXT: v_lshrrev_b32_e32 v9, 16, v25
; SI-NEXT: v_lshrrev_b32_e32 v10, 16, v26
; SI-NEXT: v_lshrrev_b32_e32 v11, 16, v27
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: v_lshlrev_b32_e32 v39, 16, v11
; SI-NEXT: v_lshlrev_b32_e32 v38, 16, v10
; SI-NEXT: v_lshlrev_b32_e32 v37, 16, v9
; SI-NEXT: v_lshlrev_b32_e32 v36, 16, v8
; SI-NEXT: v_lshlrev_b32_e32 v35, 16, v7
; SI-NEXT: v_lshlrev_b32_e32 v34, 16, v6
; SI-NEXT: v_lshlrev_b32_e32 v33, 16, v5
; SI-NEXT: v_lshlrev_b32_e32 v32, 16, v4
; SI-NEXT: v_lshlrev_b32_e32 v31, 16, v3
; SI-NEXT: v_lshlrev_b32_e32 v30, 16, v2
; SI-NEXT: v_lshlrev_b32_e32 v29, 16, v1
; SI-NEXT: v_lshlrev_b32_e32 v28, 16, v0
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB42_3
; SI-NEXT: ; %bb.1: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB42_4
; SI-NEXT: .LBB42_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB42_3: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v27
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v26
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v25
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v24
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v23
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v22
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v21
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v20
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v19
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v18
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v17
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v16
; SI-NEXT: v_or_b32_e32 v0, v0, v39
; SI-NEXT: v_or_b32_e32 v1, v1, v38
; SI-NEXT: v_or_b32_e32 v2, v2, v37
; SI-NEXT: v_or_b32_e32 v3, v3, v36
; SI-NEXT: v_or_b32_e32 v4, v4, v35
; SI-NEXT: v_or_b32_e32 v5, v5, v34
; SI-NEXT: v_or_b32_e32 v6, v6, v33
; SI-NEXT: v_or_b32_e32 v7, v7, v32
; SI-NEXT: v_or_b32_e32 v8, v8, v31
; SI-NEXT: v_or_b32_e32 v9, v9, v30
; SI-NEXT: v_or_b32_e32 v10, v10, v29
; SI-NEXT: v_or_b32_e32 v11, v11, v28
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr39
; SI-NEXT: ; implicit-def: $vgpr38
; SI-NEXT: ; implicit-def: $vgpr37
; SI-NEXT: ; implicit-def: $vgpr36
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB42_2
; SI-NEXT: .LBB42_4: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v27
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v26
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v25
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v24
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v23
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v22
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v21
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v20
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v19
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v18
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v17
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v16
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_or_b32_e32 v0, v39, v0
; SI-NEXT: v_or_b32_e32 v1, v38, v1
; SI-NEXT: v_or_b32_e32 v2, v37, v2
; SI-NEXT: v_or_b32_e32 v3, v36, v3
; SI-NEXT: v_or_b32_e32 v4, v35, v4
; SI-NEXT: v_or_b32_e32 v5, v34, v5
; SI-NEXT: v_or_b32_e32 v6, v33, v6
; SI-NEXT: v_or_b32_e32 v7, v32, v7
; SI-NEXT: v_or_b32_e32 v8, v31, v8
; SI-NEXT: v_or_b32_e32 v9, v30, v9
; SI-NEXT: v_or_b32_e32 v10, v29, v10
; SI-NEXT: v_or_b32_e32 v11, v28, v11
; SI-NEXT: v_add_i32_e32 v0, vcc, 0x30000, v0
; SI-NEXT: v_add_i32_e32 v1, vcc, 0x30000, v1
; SI-NEXT: v_add_i32_e32 v2, vcc, 0x30000, v2
; SI-NEXT: v_add_i32_e32 v3, vcc, 0x30000, v3
; SI-NEXT: v_add_i32_e32 v4, vcc, 0x30000, v4
; SI-NEXT: v_add_i32_e32 v5, vcc, 0x30000, v5
; SI-NEXT: v_add_i32_e32 v6, vcc, 0x30000, v6
; SI-NEXT: v_add_i32_e32 v7, vcc, 0x30000, v7
; SI-NEXT: v_add_i32_e32 v8, vcc, 0x30000, v8
; SI-NEXT: v_add_i32_e32 v9, vcc, 0x30000, v9
; SI-NEXT: v_add_i32_e32 v10, vcc, 0x30000, v10
; SI-NEXT: v_add_i32_e32 v11, vcc, 0x30000, v11
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v6f64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB42_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v13, 3
; VI-NEXT: v_add_u16_e32 v12, 3, v11
; VI-NEXT: v_add_u16_sdwa v11, v11, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v11, v12, v11
; VI-NEXT: v_add_u16_e32 v12, 3, v10
; VI-NEXT: v_add_u16_sdwa v10, v10, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v10, v12, v10
; VI-NEXT: v_add_u16_e32 v12, 3, v9
; VI-NEXT: v_add_u16_sdwa v9, v9, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v9, v12, v9
; VI-NEXT: v_add_u16_e32 v12, 3, v8
; VI-NEXT: v_add_u16_sdwa v8, v8, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v8, v12, v8
; VI-NEXT: v_add_u16_e32 v12, 3, v7
; VI-NEXT: v_add_u16_sdwa v7, v7, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v7, v12, v7
; VI-NEXT: v_add_u16_e32 v12, 3, v6
; VI-NEXT: v_add_u16_sdwa v6, v6, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v6, v12, v6
; VI-NEXT: v_add_u16_e32 v12, 3, v5
; VI-NEXT: v_add_u16_sdwa v5, v5, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v5, v12, v5
; VI-NEXT: v_add_u16_e32 v12, 3, v4
; VI-NEXT: v_add_u16_sdwa v4, v4, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v4, v12, v4
; VI-NEXT: v_add_u16_e32 v12, 3, v3
; VI-NEXT: v_add_u16_sdwa v3, v3, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v3, v12, v3
; VI-NEXT: v_add_u16_e32 v12, 3, v2
; VI-NEXT: v_add_u16_sdwa v2, v2, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v2, v12, v2
; VI-NEXT: v_add_u16_e32 v12, 3, v1
; VI-NEXT: v_add_u16_sdwa v1, v1, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v1, v12, v1
; VI-NEXT: v_add_u16_e32 v12, 3, v0
; VI-NEXT: v_add_u16_sdwa v0, v0, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v0, v12, v0
; VI-NEXT: .LBB42_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v6f64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB42_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB42_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v6f64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB42_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB42_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define inreg <6 x double> @bitcast_v24i16_to_v6f64_scalar(<24 x i16> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v6f64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s12, s27, 16
; SI-NEXT: s_lshr_b32 s14, s26, 16
; SI-NEXT: s_lshr_b32 s15, s25, 16
; SI-NEXT: s_lshr_b32 s13, s24, 16
; SI-NEXT: s_lshr_b32 s11, s23, 16
; SI-NEXT: s_lshr_b32 s10, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s8, s20, 16
; SI-NEXT: s_lshr_b32 s7, s19, 16
; SI-NEXT: s_lshr_b32 s6, s18, 16
; SI-NEXT: s_lshr_b32 s5, s17, 16
; SI-NEXT: s_lshr_b32 s4, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB43_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s16, s4
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_or_b32 s5, s16, s5
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s16, s6
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_or_b32 s7, s16, s7
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s16, s8
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_or_b32 s9, s16, s9
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s16, s10
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_or_b32 s11, s16, s11
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_or_b32 s13, s16, s13
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_or_b32 s14, s16, s14
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s16, s12
; SI-NEXT: s_branch .LBB43_3
; SI-NEXT: .LBB43_2: ; %cmp.true
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_or_b32 s4, s4, s16
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_or_b32 s5, s5, s16
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_or_b32 s6, s6, s16
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_or_b32 s7, s7, s16
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_or_b32 s8, s8, s16
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_or_b32 s9, s9, s16
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_or_b32 s10, s10, s16
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_or_b32 s11, s11, s16
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_or_b32 s13, s13, s16
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_or_b32 s14, s14, s16
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s12, s16
; SI-NEXT: s_add_i32 s4, s4, 0x30000
; SI-NEXT: s_add_i32 s5, s5, 0x30000
; SI-NEXT: s_add_i32 s6, s6, 0x30000
; SI-NEXT: s_add_i32 s7, s7, 0x30000
; SI-NEXT: s_add_i32 s8, s8, 0x30000
; SI-NEXT: s_add_i32 s9, s9, 0x30000
; SI-NEXT: s_add_i32 s10, s10, 0x30000
; SI-NEXT: s_add_i32 s11, s11, 0x30000
; SI-NEXT: s_add_i32 s13, s13, 0x30000
; SI-NEXT: s_add_i32 s15, s15, 0x30000
; SI-NEXT: s_add_i32 s14, s14, 0x30000
; SI-NEXT: s_add_i32 s12, s12, 0x30000
; SI-NEXT: .LBB43_3: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: v_mov_b32_e32 v2, s6
; SI-NEXT: v_mov_b32_e32 v3, s7
; SI-NEXT: v_mov_b32_e32 v4, s8
; SI-NEXT: v_mov_b32_e32 v5, s9
; SI-NEXT: v_mov_b32_e32 v6, s10
; SI-NEXT: v_mov_b32_e32 v7, s11
; SI-NEXT: v_mov_b32_e32 v8, s13
; SI-NEXT: v_mov_b32_e32 v9, s15
; SI-NEXT: v_mov_b32_e32 v10, s14
; SI-NEXT: v_mov_b32_e32 v11, s12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v6f64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB43_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s5, s27, 3
; VI-NEXT: s_and_b32 s4, s27, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s26, 3
; VI-NEXT: s_add_i32 s27, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s26, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s25, 3
; VI-NEXT: s_add_i32 s26, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s25, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s24, 3
; VI-NEXT: s_add_i32 s25, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s24, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s23, 3
; VI-NEXT: s_add_i32 s24, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s23, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s22, 3
; VI-NEXT: s_add_i32 s23, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s22, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s21, 3
; VI-NEXT: s_add_i32 s22, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s21, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s20, 3
; VI-NEXT: s_add_i32 s21, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s20, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s19, 3
; VI-NEXT: s_add_i32 s20, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s19, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s18, 3
; VI-NEXT: s_add_i32 s19, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s18, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s17, 3
; VI-NEXT: s_add_i32 s18, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s17, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s16, 3
; VI-NEXT: s_add_i32 s17, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s16, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s16, s4, 0x30000
; VI-NEXT: .LBB43_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v6f64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB43_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB43_2: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, s27, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, s26, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, s25, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, s24, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, s23, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, s22, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, s21, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, s20, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, s19, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, s18, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, s17, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, s16, 3 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v6f64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB43_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB43_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, s23, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, s22, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, s21, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, s20, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, s19, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, s18, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, s17, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, s16, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, s15, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, s14, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, s13, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, s12, 3 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define <24 x half> @bitcast_v6f64_to_v24f16(<6 x double> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v24f16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB44_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v18, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v17, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB44_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB44_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; SI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; SI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; SI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; SI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; SI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v18, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v17, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB44_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v18, 16, v18
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_or_b32_e32 v0, v0, v20
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v18
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v18, 16, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v17
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v18
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v24f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB44_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; VI-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; VI-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; VI-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; VI-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; VI-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; VI-NEXT: .LBB44_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v24f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB44_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX9-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX9-NEXT: .LBB44_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v24f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB44_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[10:11], v[10:11], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], v[8:9], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], v[6:7], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], 1.0
; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
; GFX11-NEXT: .LBB44_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define inreg <24 x half> @bitcast_v6f64_to_v24f16_scalar(<6 x double> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6f64_to_v24f16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB45_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v0, s16
; SI-NEXT: v_mov_b32_e32 v2, s18
; SI-NEXT: v_mov_b32_e32 v4, s20
; SI-NEXT: v_mov_b32_e32 v6, s22
; SI-NEXT: v_mov_b32_e32 v8, s24
; SI-NEXT: v_mov_b32_e32 v10, s26
; SI-NEXT: v_mov_b32_e32 v19, s28
; SI-NEXT: v_mov_b32_e32 v20, s29
; SI-NEXT: v_mov_b32_e32 v21, s40
; SI-NEXT: v_mov_b32_e32 v22, s41
; SI-NEXT: v_mov_b32_e32 v23, s42
; SI-NEXT: v_mov_b32_e32 v24, s43
; SI-NEXT: v_mov_b32_e32 v17, s14
; SI-NEXT: v_mov_b32_e32 v16, s12
; SI-NEXT: v_mov_b32_e32 v15, s10
; SI-NEXT: v_mov_b32_e32 v14, s8
; SI-NEXT: v_mov_b32_e32 v13, s6
; SI-NEXT: v_mov_b32_e32 v12, s4
; SI-NEXT: s_branch .LBB45_3
; SI-NEXT: .LBB45_2: ; %cmp.true
; SI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; SI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; SI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; SI-NEXT: v_lshr_b64 v[12:13], v[10:11], 16
; SI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; SI-NEXT: v_lshr_b64 v[13:14], v[8:9], 16
; SI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; SI-NEXT: v_lshr_b64 v[14:15], v[6:7], 16
; SI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; SI-NEXT: v_lshr_b64 v[15:16], v[4:5], 16
; SI-NEXT: v_lshr_b64 v[16:17], v[2:3], 16
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v20, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v24, 16, v1
; SI-NEXT: v_lshr_b64 v[17:18], v[0:1], 16
; SI-NEXT: .LBB45_3: ; %end
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v0, v0, v17
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v24
; SI-NEXT: v_or_b32_e32 v2, v2, v16
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v16, 16, v23
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v22
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v21
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v20
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v19
; SI-NEXT: v_or_b32_e32 v1, v1, v17
; SI-NEXT: v_or_b32_e32 v3, v3, v16
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6f64_to_v24f16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB45_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB45_2: ; %cmp.true
; VI-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; VI-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; VI-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; VI-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; VI-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; VI-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6f64_to_v24f16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB45_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB45_2: ; %cmp.true
; GFX9-NEXT: v_add_f64 v[10:11], s[26:27], 1.0
; GFX9-NEXT: v_add_f64 v[8:9], s[24:25], 1.0
; GFX9-NEXT: v_add_f64 v[6:7], s[22:23], 1.0
; GFX9-NEXT: v_add_f64 v[4:5], s[20:21], 1.0
; GFX9-NEXT: v_add_f64 v[2:3], s[18:19], 1.0
; GFX9-NEXT: v_add_f64 v[0:1], s[16:17], 1.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6f64_to_v24f16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB45_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB45_2: ; %cmp.true
; GFX11-NEXT: v_add_f64 v[10:11], s[22:23], 1.0
; GFX11-NEXT: v_add_f64 v[8:9], s[20:21], 1.0
; GFX11-NEXT: v_add_f64 v[6:7], s[18:19], 1.0
; GFX11-NEXT: v_add_f64 v[4:5], s[16:17], 1.0
; GFX11-NEXT: v_add_f64 v[2:3], s[14:15], 1.0
; GFX11-NEXT: v_add_f64 v[0:1], s[12:13], 1.0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <6 x double> %a, splat (double 1.000000e+00)
%a2 = bitcast <6 x double> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <6 x double> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define <6 x double> @bitcast_v24f16_to_v6f64(<24 x half> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v6f64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_mov_b32_e32 v16, v11
; SI-NEXT: v_mov_b32_e32 v17, v10
; SI-NEXT: v_mov_b32_e32 v18, v9
; SI-NEXT: v_mov_b32_e32 v19, v8
; SI-NEXT: v_mov_b32_e32 v20, v7
; SI-NEXT: v_mov_b32_e32 v21, v6
; SI-NEXT: v_mov_b32_e32 v22, v5
; SI-NEXT: v_mov_b32_e32 v23, v4
; SI-NEXT: v_mov_b32_e32 v24, v3
; SI-NEXT: v_mov_b32_e32 v25, v2
; SI-NEXT: v_mov_b32_e32 v26, v1
; SI-NEXT: v_mov_b32_e32 v27, v0
; SI-NEXT: v_lshrrev_b32_e32 v28, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v29, 16, v17
; SI-NEXT: v_lshrrev_b32_e32 v30, 16, v18
; SI-NEXT: v_lshrrev_b32_e32 v31, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v32, 16, v20
; SI-NEXT: v_lshrrev_b32_e32 v33, 16, v21
; SI-NEXT: v_lshrrev_b32_e32 v34, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v35, 16, v23
; SI-NEXT: v_lshrrev_b32_e32 v36, 16, v24
; SI-NEXT: v_lshrrev_b32_e32 v37, 16, v25
; SI-NEXT: v_lshrrev_b32_e32 v38, 16, v26
; SI-NEXT: v_lshrrev_b32_e32 v39, 16, v27
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB46_3
; SI-NEXT: ; %bb.1: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB46_4
; SI-NEXT: .LBB46_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB46_3: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v27
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v39
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v26
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v38
; SI-NEXT: v_or_b32_e32 v0, v0, v1
; SI-NEXT: v_or_b32_e32 v1, v2, v3
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v25
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v37
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v24
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v36
; SI-NEXT: v_or_b32_e32 v3, v3, v4
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v23
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v35
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v22
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v34
; SI-NEXT: v_or_b32_e32 v5, v5, v6
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v21
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v33
; SI-NEXT: v_or_b32_e32 v6, v6, v7
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v20
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v32
; SI-NEXT: v_or_b32_e32 v7, v7, v8
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v19
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v31
; SI-NEXT: v_or_b32_e32 v8, v8, v9
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v18
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v30
; SI-NEXT: v_or_b32_e32 v9, v9, v10
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v17
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v29
; SI-NEXT: v_or_b32_e32 v10, v10, v11
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v16
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v28
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr39
; SI-NEXT: ; implicit-def: $vgpr38
; SI-NEXT: ; implicit-def: $vgpr37
; SI-NEXT: ; implicit-def: $vgpr36
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB46_2
; SI-NEXT: .LBB46_4: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v0, v39
; SI-NEXT: v_cvt_f32_f16_e32 v2, v38
; SI-NEXT: v_cvt_f32_f16_e32 v1, v27
; SI-NEXT: v_cvt_f32_f16_e32 v3, v26
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT: v_or_b32_e32 v0, v1, v0
; SI-NEXT: v_or_b32_e32 v1, v3, v2
; SI-NEXT: v_cvt_f32_f16_e32 v2, v37
; SI-NEXT: v_cvt_f32_f16_e32 v3, v25
; SI-NEXT: v_cvt_f32_f16_e32 v4, v36
; SI-NEXT: v_cvt_f32_f16_e32 v5, v24
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v6, v35
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT: v_or_b32_e32 v2, v3, v2
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_or_b32_e32 v3, v5, v3
; SI-NEXT: v_cvt_f32_f16_e32 v4, v23
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, v34
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, v33
; SI-NEXT: v_cvt_f32_f16_e32 v8, v21
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_or_b32_e32 v5, v7, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, v32
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT: v_or_b32_e32 v6, v8, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, v20
; SI-NEXT: v_cvt_f32_f16_e32 v9, v31
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT: v_cvt_f32_f16_e32 v10, v19
; SI-NEXT: v_or_b32_e32 v7, v8, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v9
; SI-NEXT: v_cvt_f32_f16_e32 v9, v30
; SI-NEXT: v_cvt_f32_f16_e32 v11, v18
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_or_b32_e32 v8, v10, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, v29
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
; SI-NEXT: v_or_b32_e32 v9, v11, v9
; SI-NEXT: v_cvt_f32_f16_e32 v11, v17
; SI-NEXT: v_cvt_f32_f16_e32 v12, v28
; SI-NEXT: v_cvt_f32_f16_e32 v13, v16
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT: v_or_b32_e32 v10, v11, v10
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT: v_or_b32_e32 v11, v13, v11
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v6f64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB46_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v12, 0x200
; VI-NEXT: v_add_f16_sdwa v13, v11, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v11, 0x200, v11
; VI-NEXT: v_or_b32_e32 v11, v11, v13
; VI-NEXT: v_add_f16_sdwa v13, v10, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v10, 0x200, v10
; VI-NEXT: v_or_b32_e32 v10, v10, v13
; VI-NEXT: v_add_f16_sdwa v13, v9, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v9, 0x200, v9
; VI-NEXT: v_or_b32_e32 v9, v9, v13
; VI-NEXT: v_add_f16_sdwa v13, v8, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v8, 0x200, v8
; VI-NEXT: v_or_b32_e32 v8, v8, v13
; VI-NEXT: v_add_f16_sdwa v13, v7, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v7, 0x200, v7
; VI-NEXT: v_or_b32_e32 v7, v7, v13
; VI-NEXT: v_add_f16_sdwa v13, v6, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v6, 0x200, v6
; VI-NEXT: v_or_b32_e32 v6, v6, v13
; VI-NEXT: v_add_f16_sdwa v13, v5, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v5, 0x200, v5
; VI-NEXT: v_or_b32_e32 v5, v5, v13
; VI-NEXT: v_add_f16_sdwa v13, v4, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v4, 0x200, v4
; VI-NEXT: v_or_b32_e32 v4, v4, v13
; VI-NEXT: v_add_f16_sdwa v13, v3, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v3, 0x200, v3
; VI-NEXT: v_or_b32_e32 v3, v3, v13
; VI-NEXT: v_add_f16_sdwa v13, v2, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, 0x200, v2
; VI-NEXT: v_or_b32_e32 v2, v2, v13
; VI-NEXT: v_add_f16_sdwa v13, v1, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v1, 0x200, v1
; VI-NEXT: v_add_f16_sdwa v12, v0, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, 0x200, v0
; VI-NEXT: v_or_b32_e32 v1, v1, v13
; VI-NEXT: v_or_b32_e32 v0, v0, v12
; VI-NEXT: .LBB46_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v6f64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB46_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_movk_i32 s6, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, v11, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, v10, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, v9, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, v8, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, v7, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, v6, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, v5, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, v4, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, v3, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, v2, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, v1, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, v0, s6 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB46_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v6f64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB46_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, v11, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, v10, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, v9, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, v8, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, v7, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, v6, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, v5, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, v4, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, v3, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, v2, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, v1, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, v0, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB46_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define inreg <6 x double> @bitcast_v24f16_to_v6f64_scalar(<24 x half> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v6f64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s15, s27, 16
; SI-NEXT: s_lshr_b32 s14, s26, 16
; SI-NEXT: s_lshr_b32 s13, s25, 16
; SI-NEXT: s_lshr_b32 s12, s24, 16
; SI-NEXT: s_lshr_b32 s11, s23, 16
; SI-NEXT: s_lshr_b32 s10, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s8, s20, 16
; SI-NEXT: s_lshr_b32 s7, s19, 16
; SI-NEXT: s_lshr_b32 s6, s18, 16
; SI-NEXT: s_lshr_b32 s5, s17, 16
; SI-NEXT: s_lshr_b32 s4, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB47_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s16, s4
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_or_b32 s5, s16, s5
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s16, s6
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_or_b32 s7, s16, s7
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s16, s8
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_or_b32 s9, s16, s9
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s16, s10
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_or_b32 s11, s16, s11
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s16, s12
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_or_b32 s13, s16, s13
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_or_b32 s14, s16, s14
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: v_mov_b32_e32 v2, s6
; SI-NEXT: v_mov_b32_e32 v3, s7
; SI-NEXT: v_mov_b32_e32 v4, s8
; SI-NEXT: v_mov_b32_e32 v5, s9
; SI-NEXT: v_mov_b32_e32 v6, s10
; SI-NEXT: v_mov_b32_e32 v7, s11
; SI-NEXT: v_mov_b32_e32 v8, s12
; SI-NEXT: v_mov_b32_e32 v9, s13
; SI-NEXT: v_mov_b32_e32 v10, s14
; SI-NEXT: v_mov_b32_e32 v11, s15
; SI-NEXT: v_mov_b32_e32 v12, s16
; SI-NEXT: v_mov_b32_e32 v13, s17
; SI-NEXT: v_mov_b32_e32 v14, s18
; SI-NEXT: v_mov_b32_e32 v15, s19
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB47_2: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v0, s4
; SI-NEXT: v_cvt_f32_f16_e32 v1, s16
; SI-NEXT: v_cvt_f32_f16_e32 v2, s5
; SI-NEXT: v_cvt_f32_f16_e32 v3, s17
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_or_b32_e32 v0, v1, v0
; SI-NEXT: v_cvt_f32_f16_e32 v1, s6
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f32_f16_e32 v5, s19
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v4, v1
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
; SI-NEXT: v_cvt_f32_f16_e32 v2, s18
; SI-NEXT: v_or_b32_e32 v1, v3, v1
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_cvt_f32_f16_e32 v4, s7
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v6, s8
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_or_b32_e32 v3, v5, v3
; SI-NEXT: v_cvt_f32_f16_e32 v4, s20
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, s9
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, s21
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, s10
; SI-NEXT: v_cvt_f32_f16_e32 v8, s22
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_or_b32_e32 v5, v7, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, s11
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT: v_or_b32_e32 v6, v8, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, s23
; SI-NEXT: v_cvt_f32_f16_e32 v9, s12
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT: v_cvt_f32_f16_e32 v10, s24
; SI-NEXT: v_or_b32_e32 v7, v8, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v9
; SI-NEXT: v_cvt_f32_f16_e32 v9, s13
; SI-NEXT: v_cvt_f32_f16_e32 v11, s25
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_or_b32_e32 v8, v10, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, s14
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
; SI-NEXT: v_or_b32_e32 v9, v11, v9
; SI-NEXT: v_cvt_f32_f16_e32 v11, s26
; SI-NEXT: v_cvt_f32_f16_e32 v12, s15
; SI-NEXT: v_cvt_f32_f16_e32 v13, s27
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT: v_or_b32_e32 v10, v11, v10
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT: v_or_b32_e32 v11, v13, v11
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v6f64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB47_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB47_2: ; %cmp.true
; VI-NEXT: s_lshr_b32 s4, s27, 16
; VI-NEXT: v_mov_b32_e32 v0, 0x200
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s27, v0
; VI-NEXT: s_lshr_b32 s4, s26, 16
; VI-NEXT: v_or_b32_e32 v11, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s26, v0
; VI-NEXT: s_lshr_b32 s4, s25, 16
; VI-NEXT: v_or_b32_e32 v10, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s25, v0
; VI-NEXT: s_lshr_b32 s4, s24, 16
; VI-NEXT: v_or_b32_e32 v9, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s24, v0
; VI-NEXT: s_lshr_b32 s4, s23, 16
; VI-NEXT: v_or_b32_e32 v8, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s23, v0
; VI-NEXT: s_lshr_b32 s4, s22, 16
; VI-NEXT: v_or_b32_e32 v7, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s22, v0
; VI-NEXT: s_lshr_b32 s4, s21, 16
; VI-NEXT: v_or_b32_e32 v6, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s21, v0
; VI-NEXT: s_lshr_b32 s4, s20, 16
; VI-NEXT: v_or_b32_e32 v5, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s20, v0
; VI-NEXT: s_lshr_b32 s4, s19, 16
; VI-NEXT: v_or_b32_e32 v4, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s19, v0
; VI-NEXT: s_lshr_b32 s4, s18, 16
; VI-NEXT: v_or_b32_e32 v3, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s18, v0
; VI-NEXT: s_lshr_b32 s4, s17, 16
; VI-NEXT: v_or_b32_e32 v2, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v12, s17, v0
; VI-NEXT: s_lshr_b32 s4, s16, 16
; VI-NEXT: v_or_b32_e32 v1, v12, v1
; VI-NEXT: v_mov_b32_e32 v12, s4
; VI-NEXT: v_add_f16_sdwa v12, v12, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, s16, v0
; VI-NEXT: v_or_b32_e32 v0, v0, v12
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v6f64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB47_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB47_2: ; %cmp.true
; GFX9-NEXT: v_mov_b32_e32 v0, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, s27, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, s26, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, s25, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, s24, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, s23, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, s22, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, s21, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, s20, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, s19, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, s18, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, s17, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, s16, v0 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v6f64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB47_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB47_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, s23, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, s22, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, s21, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, s20, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, s19, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, s18, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, s17, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, s16, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, s15, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, s14, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, s13, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, s12, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <6 x double>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <6 x double>
br label %end
end:
%phi = phi <6 x double> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x double> %phi
}
define <24 x i16> @bitcast_v6i64_to_v24i16(<6 x i64> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v24i16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB48_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB48_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB48_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB48_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v0, v0, v20
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v17
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v18
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v17
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v24i16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB48_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: .LBB48_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v24i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB48_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_co_u32_e32 v10, vcc, 3, v10
; GFX9-NEXT: v_addc_co_u32_e32 v11, vcc, 0, v11, vcc
; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, 3, v8
; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, 0, v9, vcc
; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, 3, v6
; GFX9-NEXT: v_addc_co_u32_e32 v7, vcc, 0, v7, vcc
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, 3, v4
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, 3, v2
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, 3, v0
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: .LBB48_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v24i16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB48_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_co_u32 v10, vcc_lo, v10, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v11, null, 0, v11, vcc_lo
; GFX11-NEXT: v_add_co_u32 v8, vcc_lo, v8, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v9, null, 0, v9, vcc_lo
; GFX11-NEXT: v_add_co_u32 v6, vcc_lo, v6, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v7, null, 0, v7, vcc_lo
; GFX11-NEXT: v_add_co_u32 v4, vcc_lo, v4, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v5, null, 0, v5, vcc_lo
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: .LBB48_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define inreg <24 x i16> @bitcast_v6i64_to_v24i16_scalar(<6 x i64> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v24i16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc1 .LBB49_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: s_add_u32 s26, s26, 3
; SI-NEXT: s_addc_u32 s27, s27, 0
; SI-NEXT: s_add_u32 s24, s24, 3
; SI-NEXT: s_addc_u32 s25, s25, 0
; SI-NEXT: s_add_u32 s22, s22, 3
; SI-NEXT: s_addc_u32 s23, s23, 0
; SI-NEXT: s_add_u32 s20, s20, 3
; SI-NEXT: s_addc_u32 s21, s21, 0
; SI-NEXT: s_add_u32 s18, s18, 3
; SI-NEXT: s_addc_u32 s19, s19, 0
; SI-NEXT: s_add_u32 s16, s16, 3
; SI-NEXT: s_addc_u32 s17, s17, 0
; SI-NEXT: .LBB49_2: ; %cmp.false
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_and_b32 s5, s16, 0xffff
; SI-NEXT: s_lshl_b32 s7, s14, 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_or_b32 s5, s5, s7
; SI-NEXT: s_and_b32 s7, s17, 0xffff
; SI-NEXT: s_lshl_b32 s9, s43, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_or_b32 s7, s7, s9
; SI-NEXT: s_and_b32 s9, s18, 0xffff
; SI-NEXT: s_lshl_b32 s11, s12, 16
; SI-NEXT: s_or_b32 s9, s9, s11
; SI-NEXT: s_and_b32 s11, s19, 0xffff
; SI-NEXT: s_lshl_b32 s12, s42, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_or_b32 s11, s11, s12
; SI-NEXT: s_and_b32 s12, s20, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s12, s10
; SI-NEXT: s_and_b32 s12, s21, 0xffff
; SI-NEXT: s_lshl_b32 s13, s41, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_or_b32 s12, s12, s13
; SI-NEXT: s_and_b32 s13, s22, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s13, s8
; SI-NEXT: s_and_b32 s13, s23, 0xffff
; SI-NEXT: s_lshl_b32 s14, s40, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_or_b32 s13, s13, s14
; SI-NEXT: s_and_b32 s14, s24, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s14, s6
; SI-NEXT: s_and_b32 s14, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s29, 16
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_or_b32 s14, s14, s15
; SI-NEXT: s_and_b32 s15, s26, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s15, s4
; SI-NEXT: s_and_b32 s15, s27, 0xffff
; SI-NEXT: s_lshl_b32 s16, s28, 16
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: v_mov_b32_e32 v0, s5
; SI-NEXT: v_mov_b32_e32 v1, s7
; SI-NEXT: v_mov_b32_e32 v2, s9
; SI-NEXT: v_mov_b32_e32 v3, s11
; SI-NEXT: v_mov_b32_e32 v4, s10
; SI-NEXT: v_mov_b32_e32 v5, s12
; SI-NEXT: v_mov_b32_e32 v6, s8
; SI-NEXT: v_mov_b32_e32 v7, s13
; SI-NEXT: v_mov_b32_e32 v8, s6
; SI-NEXT: v_mov_b32_e32 v9, s14
; SI-NEXT: v_mov_b32_e32 v10, s4
; SI-NEXT: v_mov_b32_e32 v11, s15
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v24i16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB49_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_u32 s26, s26, 3
; VI-NEXT: s_addc_u32 s27, s27, 0
; VI-NEXT: s_add_u32 s24, s24, 3
; VI-NEXT: s_addc_u32 s25, s25, 0
; VI-NEXT: s_add_u32 s22, s22, 3
; VI-NEXT: s_addc_u32 s23, s23, 0
; VI-NEXT: s_add_u32 s20, s20, 3
; VI-NEXT: s_addc_u32 s21, s21, 0
; VI-NEXT: s_add_u32 s18, s18, 3
; VI-NEXT: s_addc_u32 s19, s19, 0
; VI-NEXT: s_add_u32 s16, s16, 3
; VI-NEXT: s_addc_u32 s17, s17, 0
; VI-NEXT: .LBB49_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v24i16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB49_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_u32 s26, s26, 3
; GFX9-NEXT: s_addc_u32 s27, s27, 0
; GFX9-NEXT: s_add_u32 s24, s24, 3
; GFX9-NEXT: s_addc_u32 s25, s25, 0
; GFX9-NEXT: s_add_u32 s22, s22, 3
; GFX9-NEXT: s_addc_u32 s23, s23, 0
; GFX9-NEXT: s_add_u32 s20, s20, 3
; GFX9-NEXT: s_addc_u32 s21, s21, 0
; GFX9-NEXT: s_add_u32 s18, s18, 3
; GFX9-NEXT: s_addc_u32 s19, s19, 0
; GFX9-NEXT: s_add_u32 s16, s16, 3
; GFX9-NEXT: s_addc_u32 s17, s17, 0
; GFX9-NEXT: .LBB49_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v24i16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB49_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_u32 s22, s22, 3
; GFX11-NEXT: s_addc_u32 s23, s23, 0
; GFX11-NEXT: s_add_u32 s20, s20, 3
; GFX11-NEXT: s_addc_u32 s21, s21, 0
; GFX11-NEXT: s_add_u32 s18, s18, 3
; GFX11-NEXT: s_addc_u32 s19, s19, 0
; GFX11-NEXT: s_add_u32 s16, s16, 3
; GFX11-NEXT: s_addc_u32 s17, s17, 0
; GFX11-NEXT: s_add_u32 s2, s2, 3
; GFX11-NEXT: s_addc_u32 s3, s3, 0
; GFX11-NEXT: s_add_u32 s0, s0, 3
; GFX11-NEXT: s_addc_u32 s1, s1, 0
; GFX11-NEXT: .LBB49_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define <6 x i64> @bitcast_v24i16_to_v6i64(<24 x i16> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v6i64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_mov_b32_e32 v16, v11
; SI-NEXT: v_mov_b32_e32 v17, v10
; SI-NEXT: v_mov_b32_e32 v18, v9
; SI-NEXT: v_mov_b32_e32 v19, v8
; SI-NEXT: v_mov_b32_e32 v20, v7
; SI-NEXT: v_mov_b32_e32 v21, v6
; SI-NEXT: v_mov_b32_e32 v22, v5
; SI-NEXT: v_mov_b32_e32 v23, v4
; SI-NEXT: v_mov_b32_e32 v24, v3
; SI-NEXT: v_mov_b32_e32 v25, v2
; SI-NEXT: v_mov_b32_e32 v26, v1
; SI-NEXT: v_mov_b32_e32 v27, v0
; SI-NEXT: v_lshrrev_b32_e32 v0, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v1, 16, v17
; SI-NEXT: v_lshrrev_b32_e32 v2, 16, v18
; SI-NEXT: v_lshrrev_b32_e32 v3, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v4, 16, v20
; SI-NEXT: v_lshrrev_b32_e32 v5, 16, v21
; SI-NEXT: v_lshrrev_b32_e32 v6, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v7, 16, v23
; SI-NEXT: v_lshrrev_b32_e32 v8, 16, v24
; SI-NEXT: v_lshrrev_b32_e32 v9, 16, v25
; SI-NEXT: v_lshrrev_b32_e32 v10, 16, v26
; SI-NEXT: v_lshrrev_b32_e32 v11, 16, v27
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: v_lshlrev_b32_e32 v39, 16, v11
; SI-NEXT: v_lshlrev_b32_e32 v38, 16, v10
; SI-NEXT: v_lshlrev_b32_e32 v37, 16, v9
; SI-NEXT: v_lshlrev_b32_e32 v36, 16, v8
; SI-NEXT: v_lshlrev_b32_e32 v35, 16, v7
; SI-NEXT: v_lshlrev_b32_e32 v34, 16, v6
; SI-NEXT: v_lshlrev_b32_e32 v33, 16, v5
; SI-NEXT: v_lshlrev_b32_e32 v32, 16, v4
; SI-NEXT: v_lshlrev_b32_e32 v31, 16, v3
; SI-NEXT: v_lshlrev_b32_e32 v30, 16, v2
; SI-NEXT: v_lshlrev_b32_e32 v29, 16, v1
; SI-NEXT: v_lshlrev_b32_e32 v28, 16, v0
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB50_3
; SI-NEXT: ; %bb.1: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB50_4
; SI-NEXT: .LBB50_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB50_3: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v27
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v26
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v25
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v24
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v23
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v22
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v21
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v20
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v19
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v18
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v17
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v16
; SI-NEXT: v_or_b32_e32 v0, v0, v39
; SI-NEXT: v_or_b32_e32 v1, v1, v38
; SI-NEXT: v_or_b32_e32 v2, v2, v37
; SI-NEXT: v_or_b32_e32 v3, v3, v36
; SI-NEXT: v_or_b32_e32 v4, v4, v35
; SI-NEXT: v_or_b32_e32 v5, v5, v34
; SI-NEXT: v_or_b32_e32 v6, v6, v33
; SI-NEXT: v_or_b32_e32 v7, v7, v32
; SI-NEXT: v_or_b32_e32 v8, v8, v31
; SI-NEXT: v_or_b32_e32 v9, v9, v30
; SI-NEXT: v_or_b32_e32 v10, v10, v29
; SI-NEXT: v_or_b32_e32 v11, v11, v28
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr39
; SI-NEXT: ; implicit-def: $vgpr38
; SI-NEXT: ; implicit-def: $vgpr37
; SI-NEXT: ; implicit-def: $vgpr36
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB50_2
; SI-NEXT: .LBB50_4: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v27
; SI-NEXT: v_add_i32_e32 v1, vcc, 3, v26
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v25
; SI-NEXT: v_add_i32_e32 v3, vcc, 3, v24
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v23
; SI-NEXT: v_add_i32_e32 v5, vcc, 3, v22
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v21
; SI-NEXT: v_add_i32_e32 v7, vcc, 3, v20
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v19
; SI-NEXT: v_add_i32_e32 v9, vcc, 3, v18
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v17
; SI-NEXT: v_add_i32_e32 v11, vcc, 3, v16
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_or_b32_e32 v0, v39, v0
; SI-NEXT: v_or_b32_e32 v1, v38, v1
; SI-NEXT: v_or_b32_e32 v2, v37, v2
; SI-NEXT: v_or_b32_e32 v3, v36, v3
; SI-NEXT: v_or_b32_e32 v4, v35, v4
; SI-NEXT: v_or_b32_e32 v5, v34, v5
; SI-NEXT: v_or_b32_e32 v6, v33, v6
; SI-NEXT: v_or_b32_e32 v7, v32, v7
; SI-NEXT: v_or_b32_e32 v8, v31, v8
; SI-NEXT: v_or_b32_e32 v9, v30, v9
; SI-NEXT: v_or_b32_e32 v10, v29, v10
; SI-NEXT: v_or_b32_e32 v11, v28, v11
; SI-NEXT: v_add_i32_e32 v0, vcc, 0x30000, v0
; SI-NEXT: v_add_i32_e32 v1, vcc, 0x30000, v1
; SI-NEXT: v_add_i32_e32 v2, vcc, 0x30000, v2
; SI-NEXT: v_add_i32_e32 v3, vcc, 0x30000, v3
; SI-NEXT: v_add_i32_e32 v4, vcc, 0x30000, v4
; SI-NEXT: v_add_i32_e32 v5, vcc, 0x30000, v5
; SI-NEXT: v_add_i32_e32 v6, vcc, 0x30000, v6
; SI-NEXT: v_add_i32_e32 v7, vcc, 0x30000, v7
; SI-NEXT: v_add_i32_e32 v8, vcc, 0x30000, v8
; SI-NEXT: v_add_i32_e32 v9, vcc, 0x30000, v9
; SI-NEXT: v_add_i32_e32 v10, vcc, 0x30000, v10
; SI-NEXT: v_add_i32_e32 v11, vcc, 0x30000, v11
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v6i64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB50_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v13, 3
; VI-NEXT: v_add_u16_e32 v12, 3, v11
; VI-NEXT: v_add_u16_sdwa v11, v11, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v11, v12, v11
; VI-NEXT: v_add_u16_e32 v12, 3, v10
; VI-NEXT: v_add_u16_sdwa v10, v10, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v10, v12, v10
; VI-NEXT: v_add_u16_e32 v12, 3, v9
; VI-NEXT: v_add_u16_sdwa v9, v9, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v9, v12, v9
; VI-NEXT: v_add_u16_e32 v12, 3, v8
; VI-NEXT: v_add_u16_sdwa v8, v8, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v8, v12, v8
; VI-NEXT: v_add_u16_e32 v12, 3, v7
; VI-NEXT: v_add_u16_sdwa v7, v7, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v7, v12, v7
; VI-NEXT: v_add_u16_e32 v12, 3, v6
; VI-NEXT: v_add_u16_sdwa v6, v6, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v6, v12, v6
; VI-NEXT: v_add_u16_e32 v12, 3, v5
; VI-NEXT: v_add_u16_sdwa v5, v5, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v5, v12, v5
; VI-NEXT: v_add_u16_e32 v12, 3, v4
; VI-NEXT: v_add_u16_sdwa v4, v4, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v4, v12, v4
; VI-NEXT: v_add_u16_e32 v12, 3, v3
; VI-NEXT: v_add_u16_sdwa v3, v3, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v3, v12, v3
; VI-NEXT: v_add_u16_e32 v12, 3, v2
; VI-NEXT: v_add_u16_sdwa v2, v2, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v2, v12, v2
; VI-NEXT: v_add_u16_e32 v12, 3, v1
; VI-NEXT: v_add_u16_sdwa v1, v1, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v1, v12, v1
; VI-NEXT: v_add_u16_e32 v12, 3, v0
; VI-NEXT: v_add_u16_sdwa v0, v0, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v0, v12, v0
; VI-NEXT: .LBB50_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v6i64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB50_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB50_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v6i64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB50_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB50_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define inreg <6 x i64> @bitcast_v24i16_to_v6i64_scalar(<24 x i16> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v6i64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s12, s27, 16
; SI-NEXT: s_lshr_b32 s14, s26, 16
; SI-NEXT: s_lshr_b32 s15, s25, 16
; SI-NEXT: s_lshr_b32 s13, s24, 16
; SI-NEXT: s_lshr_b32 s11, s23, 16
; SI-NEXT: s_lshr_b32 s10, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s8, s20, 16
; SI-NEXT: s_lshr_b32 s7, s19, 16
; SI-NEXT: s_lshr_b32 s6, s18, 16
; SI-NEXT: s_lshr_b32 s5, s17, 16
; SI-NEXT: s_lshr_b32 s4, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB51_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s16, s4
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_or_b32 s5, s16, s5
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s16, s6
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_or_b32 s7, s16, s7
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s16, s8
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_or_b32 s9, s16, s9
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s16, s10
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_or_b32 s11, s16, s11
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_or_b32 s13, s16, s13
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_or_b32 s14, s16, s14
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s16, s12
; SI-NEXT: s_branch .LBB51_3
; SI-NEXT: .LBB51_2: ; %cmp.true
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_or_b32 s4, s4, s16
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_or_b32 s5, s5, s16
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_or_b32 s6, s6, s16
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_or_b32 s7, s7, s16
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_or_b32 s8, s8, s16
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_or_b32 s9, s9, s16
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_or_b32 s10, s10, s16
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_or_b32 s11, s11, s16
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_or_b32 s13, s13, s16
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_or_b32 s14, s14, s16
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s12, s16
; SI-NEXT: s_add_i32 s4, s4, 0x30000
; SI-NEXT: s_add_i32 s5, s5, 0x30000
; SI-NEXT: s_add_i32 s6, s6, 0x30000
; SI-NEXT: s_add_i32 s7, s7, 0x30000
; SI-NEXT: s_add_i32 s8, s8, 0x30000
; SI-NEXT: s_add_i32 s9, s9, 0x30000
; SI-NEXT: s_add_i32 s10, s10, 0x30000
; SI-NEXT: s_add_i32 s11, s11, 0x30000
; SI-NEXT: s_add_i32 s13, s13, 0x30000
; SI-NEXT: s_add_i32 s15, s15, 0x30000
; SI-NEXT: s_add_i32 s14, s14, 0x30000
; SI-NEXT: s_add_i32 s12, s12, 0x30000
; SI-NEXT: .LBB51_3: ; %end
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: v_mov_b32_e32 v2, s6
; SI-NEXT: v_mov_b32_e32 v3, s7
; SI-NEXT: v_mov_b32_e32 v4, s8
; SI-NEXT: v_mov_b32_e32 v5, s9
; SI-NEXT: v_mov_b32_e32 v6, s10
; SI-NEXT: v_mov_b32_e32 v7, s11
; SI-NEXT: v_mov_b32_e32 v8, s13
; SI-NEXT: v_mov_b32_e32 v9, s15
; SI-NEXT: v_mov_b32_e32 v10, s14
; SI-NEXT: v_mov_b32_e32 v11, s12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v6i64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB51_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_i32 s5, s27, 3
; VI-NEXT: s_and_b32 s4, s27, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s26, 3
; VI-NEXT: s_add_i32 s27, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s26, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s25, 3
; VI-NEXT: s_add_i32 s26, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s25, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s24, 3
; VI-NEXT: s_add_i32 s25, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s24, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s23, 3
; VI-NEXT: s_add_i32 s24, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s23, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s22, 3
; VI-NEXT: s_add_i32 s23, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s22, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s21, 3
; VI-NEXT: s_add_i32 s22, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s21, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s20, 3
; VI-NEXT: s_add_i32 s21, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s20, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s19, 3
; VI-NEXT: s_add_i32 s20, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s19, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s18, 3
; VI-NEXT: s_add_i32 s19, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s18, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s17, 3
; VI-NEXT: s_add_i32 s18, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s17, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s5, s16, 3
; VI-NEXT: s_add_i32 s17, s4, 0x30000
; VI-NEXT: s_and_b32 s4, s16, 0xffff0000
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s16, s4, 0x30000
; VI-NEXT: .LBB51_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v6i64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB51_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB51_2: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, s27, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, s26, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, s25, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, s24, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, s23, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, s22, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, s21, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, s20, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, s19, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, s18, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, s17, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, s16, 3 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v6i64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB51_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB51_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, s23, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, s22, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, s21, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, s20, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, s19, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, s18, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, s17, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, s16, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, s15, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, s14, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, s13, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, s12, 3 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define <24 x half> @bitcast_v6i64_to_v24f16(<6 x i64> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v24f16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr15
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr14
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr13
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB52_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB52_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB52_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; SI-NEXT: v_alignbit_b32 v12, v11, v10, 16
; SI-NEXT: v_alignbit_b32 v13, v9, v8, 16
; SI-NEXT: v_alignbit_b32 v14, v7, v6, 16
; SI-NEXT: v_alignbit_b32 v15, v5, v4, 16
; SI-NEXT: v_alignbit_b32 v17, v3, v2, 16
; SI-NEXT: v_alignbit_b32 v20, v1, v0, 16
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: .LBB52_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v17
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v13
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v0, v0, v20
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v17
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v17, 16, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v15
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v14
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v14, 16, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v13
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v18
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v17
; SI-NEXT: v_or_b32_e32 v5, v5, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v24f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB52_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_add_u32_e32 v10, vcc, 3, v10
; VI-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc
; VI-NEXT: v_add_u32_e32 v8, vcc, 3, v8
; VI-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc
; VI-NEXT: v_add_u32_e32 v6, vcc, 3, v6
; VI-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc
; VI-NEXT: v_add_u32_e32 v4, vcc, 3, v4
; VI-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; VI-NEXT: v_add_u32_e32 v2, vcc, 3, v2
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT: v_add_u32_e32 v0, vcc, 3, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: .LBB52_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v24f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB52_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_add_co_u32_e32 v10, vcc, 3, v10
; GFX9-NEXT: v_addc_co_u32_e32 v11, vcc, 0, v11, vcc
; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, 3, v8
; GFX9-NEXT: v_addc_co_u32_e32 v9, vcc, 0, v9, vcc
; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, 3, v6
; GFX9-NEXT: v_addc_co_u32_e32 v7, vcc, 0, v7, vcc
; GFX9-NEXT: v_add_co_u32_e32 v4, vcc, 3, v4
; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, 3, v2
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, 3, v0
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: .LBB52_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v24f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB52_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_add_co_u32 v10, vcc_lo, v10, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v11, null, 0, v11, vcc_lo
; GFX11-NEXT: v_add_co_u32 v8, vcc_lo, v8, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v9, null, 0, v9, vcc_lo
; GFX11-NEXT: v_add_co_u32 v6, vcc_lo, v6, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v7, null, 0, v7, vcc_lo
; GFX11-NEXT: v_add_co_u32 v4, vcc_lo, v4, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v5, null, 0, v5, vcc_lo
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, 3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, 3
; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: .LBB52_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define inreg <24 x half> @bitcast_v6i64_to_v24f16_scalar(<6 x i64> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v6i64_to_v24f16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc1 .LBB53_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: s_add_u32 s26, s26, 3
; SI-NEXT: s_addc_u32 s27, s27, 0
; SI-NEXT: s_add_u32 s24, s24, 3
; SI-NEXT: s_addc_u32 s25, s25, 0
; SI-NEXT: s_add_u32 s22, s22, 3
; SI-NEXT: s_addc_u32 s23, s23, 0
; SI-NEXT: s_add_u32 s20, s20, 3
; SI-NEXT: s_addc_u32 s21, s21, 0
; SI-NEXT: s_add_u32 s18, s18, 3
; SI-NEXT: s_addc_u32 s19, s19, 0
; SI-NEXT: s_add_u32 s16, s16, 3
; SI-NEXT: s_addc_u32 s17, s17, 0
; SI-NEXT: .LBB53_2: ; %cmp.false
; SI-NEXT: s_lshr_b64 s[4:5], s[26:27], 16
; SI-NEXT: s_lshr_b64 s[6:7], s[24:25], 16
; SI-NEXT: s_lshr_b64 s[14:15], s[16:17], 16
; SI-NEXT: s_lshr_b32 s43, s17, 16
; SI-NEXT: s_lshr_b64 s[8:9], s[22:23], 16
; SI-NEXT: s_and_b32 s5, s16, 0xffff
; SI-NEXT: s_lshl_b32 s7, s14, 16
; SI-NEXT: s_lshr_b64 s[10:11], s[20:21], 16
; SI-NEXT: s_lshr_b64 s[12:13], s[18:19], 16
; SI-NEXT: s_or_b32 s5, s5, s7
; SI-NEXT: s_and_b32 s7, s17, 0xffff
; SI-NEXT: s_lshl_b32 s9, s43, 16
; SI-NEXT: s_lshr_b32 s42, s19, 16
; SI-NEXT: s_or_b32 s7, s7, s9
; SI-NEXT: s_and_b32 s9, s18, 0xffff
; SI-NEXT: s_lshl_b32 s11, s12, 16
; SI-NEXT: s_or_b32 s9, s9, s11
; SI-NEXT: s_and_b32 s11, s19, 0xffff
; SI-NEXT: s_lshl_b32 s12, s42, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_or_b32 s11, s11, s12
; SI-NEXT: s_and_b32 s12, s20, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s12, s10
; SI-NEXT: s_and_b32 s12, s21, 0xffff
; SI-NEXT: s_lshl_b32 s13, s41, 16
; SI-NEXT: s_lshr_b32 s40, s23, 16
; SI-NEXT: s_or_b32 s12, s12, s13
; SI-NEXT: s_and_b32 s13, s22, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s13, s8
; SI-NEXT: s_and_b32 s13, s23, 0xffff
; SI-NEXT: s_lshl_b32 s14, s40, 16
; SI-NEXT: s_lshr_b32 s29, s25, 16
; SI-NEXT: s_or_b32 s13, s13, s14
; SI-NEXT: s_and_b32 s14, s24, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s14, s6
; SI-NEXT: s_and_b32 s14, s25, 0xffff
; SI-NEXT: s_lshl_b32 s15, s29, 16
; SI-NEXT: s_lshr_b32 s28, s27, 16
; SI-NEXT: s_or_b32 s14, s14, s15
; SI-NEXT: s_and_b32 s15, s26, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s15, s4
; SI-NEXT: s_and_b32 s15, s27, 0xffff
; SI-NEXT: s_lshl_b32 s16, s28, 16
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: v_mov_b32_e32 v0, s5
; SI-NEXT: v_mov_b32_e32 v1, s7
; SI-NEXT: v_mov_b32_e32 v2, s9
; SI-NEXT: v_mov_b32_e32 v3, s11
; SI-NEXT: v_mov_b32_e32 v4, s10
; SI-NEXT: v_mov_b32_e32 v5, s12
; SI-NEXT: v_mov_b32_e32 v6, s8
; SI-NEXT: v_mov_b32_e32 v7, s13
; SI-NEXT: v_mov_b32_e32 v8, s6
; SI-NEXT: v_mov_b32_e32 v9, s14
; SI-NEXT: v_mov_b32_e32 v10, s4
; SI-NEXT: v_mov_b32_e32 v11, s15
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v6i64_to_v24f16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB53_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_add_u32 s26, s26, 3
; VI-NEXT: s_addc_u32 s27, s27, 0
; VI-NEXT: s_add_u32 s24, s24, 3
; VI-NEXT: s_addc_u32 s25, s25, 0
; VI-NEXT: s_add_u32 s22, s22, 3
; VI-NEXT: s_addc_u32 s23, s23, 0
; VI-NEXT: s_add_u32 s20, s20, 3
; VI-NEXT: s_addc_u32 s21, s21, 0
; VI-NEXT: s_add_u32 s18, s18, 3
; VI-NEXT: s_addc_u32 s19, s19, 0
; VI-NEXT: s_add_u32 s16, s16, 3
; VI-NEXT: s_addc_u32 s17, s17, 0
; VI-NEXT: .LBB53_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v6i64_to_v24f16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB53_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_add_u32 s26, s26, 3
; GFX9-NEXT: s_addc_u32 s27, s27, 0
; GFX9-NEXT: s_add_u32 s24, s24, 3
; GFX9-NEXT: s_addc_u32 s25, s25, 0
; GFX9-NEXT: s_add_u32 s22, s22, 3
; GFX9-NEXT: s_addc_u32 s23, s23, 0
; GFX9-NEXT: s_add_u32 s20, s20, 3
; GFX9-NEXT: s_addc_u32 s21, s21, 0
; GFX9-NEXT: s_add_u32 s18, s18, 3
; GFX9-NEXT: s_addc_u32 s19, s19, 0
; GFX9-NEXT: s_add_u32 s16, s16, 3
; GFX9-NEXT: s_addc_u32 s17, s17, 0
; GFX9-NEXT: .LBB53_2: ; %end
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v6i64_to_v24f16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc1 .LBB53_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: s_add_u32 s22, s22, 3
; GFX11-NEXT: s_addc_u32 s23, s23, 0
; GFX11-NEXT: s_add_u32 s20, s20, 3
; GFX11-NEXT: s_addc_u32 s21, s21, 0
; GFX11-NEXT: s_add_u32 s18, s18, 3
; GFX11-NEXT: s_addc_u32 s19, s19, 0
; GFX11-NEXT: s_add_u32 s16, s16, 3
; GFX11-NEXT: s_addc_u32 s17, s17, 0
; GFX11-NEXT: s_add_u32 s2, s2, 3
; GFX11-NEXT: s_addc_u32 s3, s3, 0
; GFX11-NEXT: s_add_u32 s0, s0, 3
; GFX11-NEXT: s_addc_u32 s1, s1, 0
; GFX11-NEXT: .LBB53_2: ; %end
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <6 x i64> %a, splat (i64 3)
%a2 = bitcast <6 x i64> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <6 x i64> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define <6 x i64> @bitcast_v24f16_to_v6i64(<24 x half> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v6i64:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_mov_b32_e32 v16, v11
; SI-NEXT: v_mov_b32_e32 v17, v10
; SI-NEXT: v_mov_b32_e32 v18, v9
; SI-NEXT: v_mov_b32_e32 v19, v8
; SI-NEXT: v_mov_b32_e32 v20, v7
; SI-NEXT: v_mov_b32_e32 v21, v6
; SI-NEXT: v_mov_b32_e32 v22, v5
; SI-NEXT: v_mov_b32_e32 v23, v4
; SI-NEXT: v_mov_b32_e32 v24, v3
; SI-NEXT: v_mov_b32_e32 v25, v2
; SI-NEXT: v_mov_b32_e32 v26, v1
; SI-NEXT: v_mov_b32_e32 v27, v0
; SI-NEXT: v_lshrrev_b32_e32 v28, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v29, 16, v17
; SI-NEXT: v_lshrrev_b32_e32 v30, 16, v18
; SI-NEXT: v_lshrrev_b32_e32 v31, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v32, 16, v20
; SI-NEXT: v_lshrrev_b32_e32 v33, 16, v21
; SI-NEXT: v_lshrrev_b32_e32 v34, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v35, 16, v23
; SI-NEXT: v_lshrrev_b32_e32 v36, 16, v24
; SI-NEXT: v_lshrrev_b32_e32 v37, 16, v25
; SI-NEXT: v_lshrrev_b32_e32 v38, 16, v26
; SI-NEXT: v_lshrrev_b32_e32 v39, 16, v27
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB54_3
; SI-NEXT: ; %bb.1: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execnz .LBB54_4
; SI-NEXT: .LBB54_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB54_3: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v27
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v39
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v26
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v38
; SI-NEXT: v_or_b32_e32 v0, v0, v1
; SI-NEXT: v_or_b32_e32 v1, v2, v3
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v25
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v37
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v24
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v36
; SI-NEXT: v_or_b32_e32 v3, v3, v4
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v23
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v35
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v22
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v34
; SI-NEXT: v_or_b32_e32 v5, v5, v6
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v21
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v33
; SI-NEXT: v_or_b32_e32 v6, v6, v7
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v20
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v32
; SI-NEXT: v_or_b32_e32 v7, v7, v8
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v19
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v31
; SI-NEXT: v_or_b32_e32 v8, v8, v9
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v18
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v30
; SI-NEXT: v_or_b32_e32 v9, v9, v10
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v17
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v29
; SI-NEXT: v_or_b32_e32 v10, v10, v11
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v16
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v28
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr23
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr20
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr17
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr39
; SI-NEXT: ; implicit-def: $vgpr38
; SI-NEXT: ; implicit-def: $vgpr37
; SI-NEXT: ; implicit-def: $vgpr36
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB54_2
; SI-NEXT: .LBB54_4: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v0, v39
; SI-NEXT: v_cvt_f32_f16_e32 v2, v38
; SI-NEXT: v_cvt_f32_f16_e32 v1, v27
; SI-NEXT: v_cvt_f32_f16_e32 v3, v26
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT: v_or_b32_e32 v0, v1, v0
; SI-NEXT: v_or_b32_e32 v1, v3, v2
; SI-NEXT: v_cvt_f32_f16_e32 v2, v37
; SI-NEXT: v_cvt_f32_f16_e32 v3, v25
; SI-NEXT: v_cvt_f32_f16_e32 v4, v36
; SI-NEXT: v_cvt_f32_f16_e32 v5, v24
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v6, v35
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT: v_or_b32_e32 v2, v3, v2
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_or_b32_e32 v3, v5, v3
; SI-NEXT: v_cvt_f32_f16_e32 v4, v23
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, v34
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, v33
; SI-NEXT: v_cvt_f32_f16_e32 v8, v21
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_or_b32_e32 v5, v7, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, v32
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT: v_or_b32_e32 v6, v8, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, v20
; SI-NEXT: v_cvt_f32_f16_e32 v9, v31
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT: v_cvt_f32_f16_e32 v10, v19
; SI-NEXT: v_or_b32_e32 v7, v8, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v9
; SI-NEXT: v_cvt_f32_f16_e32 v9, v30
; SI-NEXT: v_cvt_f32_f16_e32 v11, v18
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_or_b32_e32 v8, v10, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, v29
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
; SI-NEXT: v_or_b32_e32 v9, v11, v9
; SI-NEXT: v_cvt_f32_f16_e32 v11, v17
; SI-NEXT: v_cvt_f32_f16_e32 v12, v28
; SI-NEXT: v_cvt_f32_f16_e32 v13, v16
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT: v_or_b32_e32 v10, v11, v10
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT: v_or_b32_e32 v11, v13, v11
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v6i64:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB54_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v12, 0x200
; VI-NEXT: v_add_f16_sdwa v13, v11, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v11, 0x200, v11
; VI-NEXT: v_or_b32_e32 v11, v11, v13
; VI-NEXT: v_add_f16_sdwa v13, v10, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v10, 0x200, v10
; VI-NEXT: v_or_b32_e32 v10, v10, v13
; VI-NEXT: v_add_f16_sdwa v13, v9, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v9, 0x200, v9
; VI-NEXT: v_or_b32_e32 v9, v9, v13
; VI-NEXT: v_add_f16_sdwa v13, v8, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v8, 0x200, v8
; VI-NEXT: v_or_b32_e32 v8, v8, v13
; VI-NEXT: v_add_f16_sdwa v13, v7, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v7, 0x200, v7
; VI-NEXT: v_or_b32_e32 v7, v7, v13
; VI-NEXT: v_add_f16_sdwa v13, v6, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v6, 0x200, v6
; VI-NEXT: v_or_b32_e32 v6, v6, v13
; VI-NEXT: v_add_f16_sdwa v13, v5, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v5, 0x200, v5
; VI-NEXT: v_or_b32_e32 v5, v5, v13
; VI-NEXT: v_add_f16_sdwa v13, v4, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v4, 0x200, v4
; VI-NEXT: v_or_b32_e32 v4, v4, v13
; VI-NEXT: v_add_f16_sdwa v13, v3, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v3, 0x200, v3
; VI-NEXT: v_or_b32_e32 v3, v3, v13
; VI-NEXT: v_add_f16_sdwa v13, v2, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, 0x200, v2
; VI-NEXT: v_or_b32_e32 v2, v2, v13
; VI-NEXT: v_add_f16_sdwa v13, v1, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v1, 0x200, v1
; VI-NEXT: v_add_f16_sdwa v12, v0, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, 0x200, v0
; VI-NEXT: v_or_b32_e32 v1, v1, v13
; VI-NEXT: v_or_b32_e32 v0, v0, v12
; VI-NEXT: .LBB54_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v6i64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB54_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_movk_i32 s6, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, v11, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, v10, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, v9, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, v8, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, v7, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, v6, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, v5, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, v4, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, v3, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, v2, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, v1, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, v0, s6 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB54_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v6i64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB54_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, v11, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, v10, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, v9, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, v8, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, v7, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, v6, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, v5, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, v4, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, v3, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, v2, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, v1, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, v0, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB54_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define inreg <6 x i64> @bitcast_v24f16_to_v6i64_scalar(<24 x half> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v6i64_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s15, s27, 16
; SI-NEXT: s_lshr_b32 s14, s26, 16
; SI-NEXT: s_lshr_b32 s13, s25, 16
; SI-NEXT: s_lshr_b32 s12, s24, 16
; SI-NEXT: s_lshr_b32 s11, s23, 16
; SI-NEXT: s_lshr_b32 s10, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s8, s20, 16
; SI-NEXT: s_lshr_b32 s7, s19, 16
; SI-NEXT: s_lshr_b32 s6, s18, 16
; SI-NEXT: s_lshr_b32 s5, s17, 16
; SI-NEXT: s_lshr_b32 s4, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB55_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s16, s16, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_or_b32 s4, s16, s4
; SI-NEXT: s_and_b32 s16, s17, 0xffff
; SI-NEXT: s_lshl_b32 s5, s5, 16
; SI-NEXT: s_or_b32 s5, s16, s5
; SI-NEXT: s_and_b32 s16, s18, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_or_b32 s6, s16, s6
; SI-NEXT: s_and_b32 s16, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s7, 16
; SI-NEXT: s_or_b32 s7, s16, s7
; SI-NEXT: s_and_b32 s16, s20, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_or_b32 s8, s16, s8
; SI-NEXT: s_and_b32 s16, s21, 0xffff
; SI-NEXT: s_lshl_b32 s9, s9, 16
; SI-NEXT: s_or_b32 s9, s16, s9
; SI-NEXT: s_and_b32 s16, s22, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_or_b32 s10, s16, s10
; SI-NEXT: s_and_b32 s16, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s11, 16
; SI-NEXT: s_or_b32 s11, s16, s11
; SI-NEXT: s_and_b32 s16, s24, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_or_b32 s12, s16, s12
; SI-NEXT: s_and_b32 s16, s25, 0xffff
; SI-NEXT: s_lshl_b32 s13, s13, 16
; SI-NEXT: s_or_b32 s13, s16, s13
; SI-NEXT: s_and_b32 s16, s26, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_or_b32 s14, s16, s14
; SI-NEXT: s_and_b32 s16, s27, 0xffff
; SI-NEXT: s_lshl_b32 s15, s15, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: v_mov_b32_e32 v2, s6
; SI-NEXT: v_mov_b32_e32 v3, s7
; SI-NEXT: v_mov_b32_e32 v4, s8
; SI-NEXT: v_mov_b32_e32 v5, s9
; SI-NEXT: v_mov_b32_e32 v6, s10
; SI-NEXT: v_mov_b32_e32 v7, s11
; SI-NEXT: v_mov_b32_e32 v8, s12
; SI-NEXT: v_mov_b32_e32 v9, s13
; SI-NEXT: v_mov_b32_e32 v10, s14
; SI-NEXT: v_mov_b32_e32 v11, s15
; SI-NEXT: v_mov_b32_e32 v12, s16
; SI-NEXT: v_mov_b32_e32 v13, s17
; SI-NEXT: v_mov_b32_e32 v14, s18
; SI-NEXT: v_mov_b32_e32 v15, s19
; SI-NEXT: s_setpc_b64 s[30:31]
; SI-NEXT: .LBB55_2: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v0, s4
; SI-NEXT: v_cvt_f32_f16_e32 v1, s16
; SI-NEXT: v_cvt_f32_f16_e32 v2, s5
; SI-NEXT: v_cvt_f32_f16_e32 v3, s17
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_or_b32_e32 v0, v1, v0
; SI-NEXT: v_cvt_f32_f16_e32 v1, s6
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f32_f16_e32 v5, s19
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v4, v1
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v2
; SI-NEXT: v_cvt_f32_f16_e32 v2, s18
; SI-NEXT: v_or_b32_e32 v1, v3, v1
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_cvt_f32_f16_e32 v4, s7
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v6, s8
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; SI-NEXT: v_or_b32_e32 v3, v5, v3
; SI-NEXT: v_cvt_f32_f16_e32 v4, s20
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, s9
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, s21
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
; SI-NEXT: v_cvt_f32_f16_e32 v6, s10
; SI-NEXT: v_cvt_f32_f16_e32 v8, s22
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_or_b32_e32 v5, v7, v5
; SI-NEXT: v_cvt_f32_f16_e32 v7, s11
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT: v_or_b32_e32 v6, v8, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, s23
; SI-NEXT: v_cvt_f32_f16_e32 v9, s12
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT: v_cvt_f32_f16_e32 v10, s24
; SI-NEXT: v_or_b32_e32 v7, v8, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v9
; SI-NEXT: v_cvt_f32_f16_e32 v9, s13
; SI-NEXT: v_cvt_f32_f16_e32 v11, s25
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_or_b32_e32 v8, v10, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, s14
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v9
; SI-NEXT: v_or_b32_e32 v9, v11, v9
; SI-NEXT: v_cvt_f32_f16_e32 v11, s26
; SI-NEXT: v_cvt_f32_f16_e32 v12, s15
; SI-NEXT: v_cvt_f32_f16_e32 v13, s27
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT: v_or_b32_e32 v10, v11, v10
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT: v_or_b32_e32 v11, v13, v11
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v6i64_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB55_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB55_2: ; %cmp.true
; VI-NEXT: s_lshr_b32 s4, s27, 16
; VI-NEXT: v_mov_b32_e32 v0, 0x200
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s27, v0
; VI-NEXT: s_lshr_b32 s4, s26, 16
; VI-NEXT: v_or_b32_e32 v11, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s26, v0
; VI-NEXT: s_lshr_b32 s4, s25, 16
; VI-NEXT: v_or_b32_e32 v10, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s25, v0
; VI-NEXT: s_lshr_b32 s4, s24, 16
; VI-NEXT: v_or_b32_e32 v9, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s24, v0
; VI-NEXT: s_lshr_b32 s4, s23, 16
; VI-NEXT: v_or_b32_e32 v8, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s23, v0
; VI-NEXT: s_lshr_b32 s4, s22, 16
; VI-NEXT: v_or_b32_e32 v7, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s22, v0
; VI-NEXT: s_lshr_b32 s4, s21, 16
; VI-NEXT: v_or_b32_e32 v6, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s21, v0
; VI-NEXT: s_lshr_b32 s4, s20, 16
; VI-NEXT: v_or_b32_e32 v5, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s20, v0
; VI-NEXT: s_lshr_b32 s4, s19, 16
; VI-NEXT: v_or_b32_e32 v4, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s19, v0
; VI-NEXT: s_lshr_b32 s4, s18, 16
; VI-NEXT: v_or_b32_e32 v3, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v2, s18, v0
; VI-NEXT: s_lshr_b32 s4, s17, 16
; VI-NEXT: v_or_b32_e32 v2, v2, v1
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v12, s17, v0
; VI-NEXT: s_lshr_b32 s4, s16, 16
; VI-NEXT: v_or_b32_e32 v1, v12, v1
; VI-NEXT: v_mov_b32_e32 v12, s4
; VI-NEXT: v_add_f16_sdwa v12, v12, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, s16, v0
; VI-NEXT: v_or_b32_e32 v0, v0, v12
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v6i64_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB55_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB55_2: ; %cmp.true
; GFX9-NEXT: v_mov_b32_e32 v0, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, s27, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, s26, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, s25, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, s24, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, s23, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, s22, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, s21, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, s20, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, s19, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, s18, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, s17, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, s16, v0 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v6i64_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB55_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB55_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, s23, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, s22, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, s21, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, s20, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, s19, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, s18, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, s17, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, s16, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, s15, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, s14, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, s13, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, s12, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <6 x i64>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <6 x i64>
br label %end
end:
%phi = phi <6 x i64> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <6 x i64> %phi
}
define <24 x half> @bitcast_v24i16_to_v24f16(<24 x i16> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v24f16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_lshrrev_b32_e32 v13, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v10
; SI-NEXT: v_lshrrev_b32_e32 v14, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v8
; SI-NEXT: v_lshrrev_b32_e32 v15, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v6
; SI-NEXT: v_lshrrev_b32_e32 v17, 16, v5
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v4
; SI-NEXT: v_lshrrev_b32_e32 v20, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v2
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v1
; SI-NEXT: v_lshrrev_b32_e32 v24, 16, v0
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: v_lshlrev_b32_e32 v38, 16, v23
; SI-NEXT: v_lshlrev_b32_e32 v36, 16, v24
; SI-NEXT: v_lshlrev_b32_e32 v48, 16, v20
; SI-NEXT: v_lshlrev_b32_e32 v37, 16, v22
; SI-NEXT: v_lshlrev_b32_e32 v49, 16, v17
; SI-NEXT: v_lshlrev_b32_e32 v39, 16, v21
; SI-NEXT: v_lshlrev_b32_e32 v51, 16, v15
; SI-NEXT: v_lshlrev_b32_e32 v50, 16, v19
; SI-NEXT: v_lshlrev_b32_e32 v53, 16, v14
; SI-NEXT: v_lshlrev_b32_e32 v52, 16, v18
; SI-NEXT: v_lshlrev_b32_e32 v55, 16, v13
; SI-NEXT: v_lshlrev_b32_e32 v54, 16, v16
; SI-NEXT: ; implicit-def: $vgpr35
; SI-NEXT: ; implicit-def: $vgpr28
; SI-NEXT: ; implicit-def: $vgpr27
; SI-NEXT: ; implicit-def: $vgpr30
; SI-NEXT: ; implicit-def: $vgpr29
; SI-NEXT: ; implicit-def: $vgpr25
; SI-NEXT: ; implicit-def: $vgpr26
; SI-NEXT: ; implicit-def: $vgpr31
; SI-NEXT: ; implicit-def: $vgpr22
; SI-NEXT: ; implicit-def: $vgpr24
; SI-NEXT: ; implicit-def: $vgpr32
; SI-NEXT: ; implicit-def: $vgpr19
; SI-NEXT: ; implicit-def: $vgpr21
; SI-NEXT: ; implicit-def: $vgpr33
; SI-NEXT: ; implicit-def: $vgpr16
; SI-NEXT: ; implicit-def: $vgpr18
; SI-NEXT: ; implicit-def: $vgpr34
; SI-NEXT: ; implicit-def: $vgpr12
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_cbranch_execz .LBB56_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_or_b32_e32 v27, v1, v38
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v3
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_or_b32_e32 v25, v1, v48
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v5
; SI-NEXT: v_or_b32_e32 v35, v0, v36
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v2
; SI-NEXT: v_or_b32_e32 v22, v1, v49
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v7
; SI-NEXT: v_or_b32_e32 v30, v0, v37
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v4
; SI-NEXT: v_or_b32_e32 v19, v1, v51
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v9
; SI-NEXT: v_or_b32_e32 v26, v0, v39
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v6
; SI-NEXT: v_or_b32_e32 v16, v1, v53
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v11
; SI-NEXT: v_or_b32_e32 v24, v0, v50
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v8
; SI-NEXT: v_or_b32_e32 v12, v1, v55
; SI-NEXT: v_or_b32_e32 v21, v0, v52
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v10
; SI-NEXT: v_alignbit_b32 v28, v27, v36, 16
; SI-NEXT: v_alignbit_b32 v29, v25, v37, 16
; SI-NEXT: v_alignbit_b32 v31, v22, v39, 16
; SI-NEXT: v_alignbit_b32 v32, v19, v50, 16
; SI-NEXT: v_alignbit_b32 v33, v16, v52, 16
; SI-NEXT: v_alignbit_b32 v34, v12, v54, 16
; SI-NEXT: v_or_b32_e32 v18, v0, v54
; SI-NEXT: ; implicit-def: $vgpr0
; SI-NEXT: ; implicit-def: $vgpr1
; SI-NEXT: ; implicit-def: $vgpr2
; SI-NEXT: ; implicit-def: $vgpr3
; SI-NEXT: ; implicit-def: $vgpr4
; SI-NEXT: ; implicit-def: $vgpr5
; SI-NEXT: ; implicit-def: $vgpr6
; SI-NEXT: ; implicit-def: $vgpr7
; SI-NEXT: ; implicit-def: $vgpr8
; SI-NEXT: ; implicit-def: $vgpr9
; SI-NEXT: ; implicit-def: $vgpr10
; SI-NEXT: ; implicit-def: $vgpr11
; SI-NEXT: ; implicit-def: $vgpr38
; SI-NEXT: ; implicit-def: $vgpr36
; SI-NEXT: ; implicit-def: $vgpr48
; SI-NEXT: ; implicit-def: $vgpr37
; SI-NEXT: ; implicit-def: $vgpr49
; SI-NEXT: ; implicit-def: $vgpr39
; SI-NEXT: ; implicit-def: $vgpr51
; SI-NEXT: ; implicit-def: $vgpr50
; SI-NEXT: ; implicit-def: $vgpr53
; SI-NEXT: ; implicit-def: $vgpr52
; SI-NEXT: ; implicit-def: $vgpr55
; SI-NEXT: ; implicit-def: $vgpr54
; SI-NEXT: .LBB56_2: ; %Flow
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB56_4
; SI-NEXT: ; %bb.3: ; %cmp.true
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v10
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v8
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v6
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v4
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v2
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v0
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_or_b32_e32 v10, v54, v10
; SI-NEXT: v_or_b32_e32 v8, v52, v8
; SI-NEXT: v_or_b32_e32 v6, v50, v6
; SI-NEXT: v_or_b32_e32 v4, v39, v4
; SI-NEXT: v_or_b32_e32 v2, v37, v2
; SI-NEXT: v_or_b32_e32 v0, v36, v0
; SI-NEXT: v_add_i32_e32 v18, vcc, 0x30000, v10
; SI-NEXT: v_add_i32_e32 v10, vcc, 3, v11
; SI-NEXT: v_add_i32_e32 v21, vcc, 0x30000, v8
; SI-NEXT: v_add_i32_e32 v8, vcc, 3, v9
; SI-NEXT: v_add_i32_e32 v24, vcc, 0x30000, v6
; SI-NEXT: v_add_i32_e32 v6, vcc, 3, v7
; SI-NEXT: v_add_i32_e32 v26, vcc, 0x30000, v4
; SI-NEXT: v_add_i32_e32 v4, vcc, 3, v5
; SI-NEXT: v_add_i32_e32 v30, vcc, 0x30000, v2
; SI-NEXT: v_add_i32_e32 v2, vcc, 3, v3
; SI-NEXT: v_add_i32_e32 v35, vcc, 0x30000, v0
; SI-NEXT: v_add_i32_e32 v0, vcc, 3, v1
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_or_b32_e32 v10, v55, v10
; SI-NEXT: v_or_b32_e32 v8, v53, v8
; SI-NEXT: v_or_b32_e32 v6, v51, v6
; SI-NEXT: v_or_b32_e32 v4, v49, v4
; SI-NEXT: v_or_b32_e32 v2, v48, v2
; SI-NEXT: v_or_b32_e32 v0, v38, v0
; SI-NEXT: v_add_i32_e32 v12, vcc, 0x30000, v10
; SI-NEXT: v_add_i32_e32 v16, vcc, 0x30000, v8
; SI-NEXT: v_add_i32_e32 v19, vcc, 0x30000, v6
; SI-NEXT: v_add_i32_e32 v22, vcc, 0x30000, v4
; SI-NEXT: v_add_i32_e32 v25, vcc, 0x30000, v2
; SI-NEXT: v_add_i32_e32 v27, vcc, 0x30000, v0
; SI-NEXT: v_alignbit_b32 v28, v27, v35, 16
; SI-NEXT: v_alignbit_b32 v29, v25, v30, 16
; SI-NEXT: v_alignbit_b32 v31, v22, v26, 16
; SI-NEXT: v_alignbit_b32 v32, v19, v24, 16
; SI-NEXT: v_alignbit_b32 v33, v16, v21, 16
; SI-NEXT: v_alignbit_b32 v34, v12, v18, 16
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v27
; SI-NEXT: v_lshrrev_b32_e32 v20, 16, v25
; SI-NEXT: v_lshrrev_b32_e32 v17, 16, v22
; SI-NEXT: v_lshrrev_b32_e32 v15, 16, v19
; SI-NEXT: v_lshrrev_b32_e32 v14, 16, v16
; SI-NEXT: v_lshrrev_b32_e32 v13, 16, v12
; SI-NEXT: .LBB56_4: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v35
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v28
; SI-NEXT: v_or_b32_e32 v0, v0, v1
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v27
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v23
; SI-NEXT: v_or_b32_e32 v1, v1, v2
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v30
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v29
; SI-NEXT: v_or_b32_e32 v2, v2, v3
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v25
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v20
; SI-NEXT: v_or_b32_e32 v3, v3, v4
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v26
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v31
; SI-NEXT: v_or_b32_e32 v4, v4, v5
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v22
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v17
; SI-NEXT: v_or_b32_e32 v5, v5, v6
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v24
; SI-NEXT: v_lshlrev_b32_e32 v7, 16, v32
; SI-NEXT: v_or_b32_e32 v6, v6, v7
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v19
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v15
; SI-NEXT: v_or_b32_e32 v7, v7, v8
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v21
; SI-NEXT: v_lshlrev_b32_e32 v9, 16, v33
; SI-NEXT: v_or_b32_e32 v8, v8, v9
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v16
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v14
; SI-NEXT: v_or_b32_e32 v9, v9, v10
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v18
; SI-NEXT: v_lshlrev_b32_e32 v11, 16, v34
; SI-NEXT: v_or_b32_e32 v10, v10, v11
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v12
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v13
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v24f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB56_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v12, 3
; VI-NEXT: v_add_u16_sdwa v19, v11, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_e32 v11, 3, v11
; VI-NEXT: v_or_b32_e32 v11, v11, v19
; VI-NEXT: v_add_u16_sdwa v19, v10, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_e32 v10, 3, v10
; VI-NEXT: v_or_b32_e32 v10, v10, v19
; VI-NEXT: v_add_u16_sdwa v19, v9, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_e32 v9, 3, v9
; VI-NEXT: v_or_b32_e32 v9, v9, v19
; VI-NEXT: v_add_u16_sdwa v19, v8, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_e32 v8, 3, v8
; VI-NEXT: v_add_u16_sdwa v13, v0, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_sdwa v14, v1, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_sdwa v15, v2, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_sdwa v16, v3, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_sdwa v17, v4, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_sdwa v18, v5, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v8, v8, v19
; VI-NEXT: v_add_u16_sdwa v19, v6, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_sdwa v12, v7, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_u16_e32 v7, 3, v7
; VI-NEXT: v_add_u16_e32 v6, 3, v6
; VI-NEXT: v_add_u16_e32 v5, 3, v5
; VI-NEXT: v_add_u16_e32 v4, 3, v4
; VI-NEXT: v_add_u16_e32 v3, 3, v3
; VI-NEXT: v_add_u16_e32 v2, 3, v2
; VI-NEXT: v_add_u16_e32 v1, 3, v1
; VI-NEXT: v_add_u16_e32 v0, 3, v0
; VI-NEXT: v_or_b32_e32 v7, v7, v12
; VI-NEXT: v_or_b32_e32 v6, v6, v19
; VI-NEXT: v_or_b32_e32 v5, v5, v18
; VI-NEXT: v_or_b32_e32 v4, v4, v17
; VI-NEXT: v_or_b32_e32 v3, v3, v16
; VI-NEXT: v_or_b32_e32 v2, v2, v15
; VI-NEXT: v_or_b32_e32 v1, v1, v14
; VI-NEXT: v_or_b32_e32 v0, v0, v13
; VI-NEXT: .LBB56_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v24f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB56_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB56_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v24f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB56_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, v11, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, v10, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, v9, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, v8, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, v7, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, v6, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, v5, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, v4, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, v3, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, v2, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, v1, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, v0, 3 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB56_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define inreg <24 x half> @bitcast_v24i16_to_v24f16_scalar(<24 x i16> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24i16_to_v24f16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s44, s27, 16
; SI-NEXT: s_lshr_b32 s4, s26, 16
; SI-NEXT: s_lshr_b32 s43, s25, 16
; SI-NEXT: s_lshr_b32 s6, s24, 16
; SI-NEXT: s_lshr_b32 s42, s23, 16
; SI-NEXT: s_lshr_b32 s8, s22, 16
; SI-NEXT: s_lshr_b32 s41, s21, 16
; SI-NEXT: s_lshr_b32 s10, s20, 16
; SI-NEXT: s_lshr_b32 s40, s19, 16
; SI-NEXT: s_lshr_b32 s12, s18, 16
; SI-NEXT: s_lshr_b32 s29, s17, 16
; SI-NEXT: s_lshr_b32 s14, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB57_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: s_and_b32 s5, s16, 0xffff
; SI-NEXT: s_lshl_b32 s46, s14, 16
; SI-NEXT: s_or_b32 s14, s5, s46
; SI-NEXT: s_and_b32 s5, s18, 0xffff
; SI-NEXT: s_lshl_b32 s56, s12, 16
; SI-NEXT: s_or_b32 s12, s5, s56
; SI-NEXT: s_and_b32 s5, s20, 0xffff
; SI-NEXT: s_lshl_b32 s58, s10, 16
; SI-NEXT: s_or_b32 s10, s5, s58
; SI-NEXT: s_and_b32 s5, s22, 0xffff
; SI-NEXT: s_lshl_b32 s60, s8, 16
; SI-NEXT: s_or_b32 s8, s5, s60
; SI-NEXT: s_and_b32 s5, s24, 0xffff
; SI-NEXT: s_lshl_b32 s62, s6, 16
; SI-NEXT: s_or_b32 s6, s5, s62
; SI-NEXT: s_and_b32 s5, s26, 0xffff
; SI-NEXT: s_lshl_b32 s72, s4, 16
; SI-NEXT: s_or_b32 s4, s5, s72
; SI-NEXT: s_and_b32 s5, s17, 0xffff
; SI-NEXT: s_lshl_b32 s7, s29, 16
; SI-NEXT: s_or_b32 s47, s5, s7
; SI-NEXT: s_and_b32 s5, s19, 0xffff
; SI-NEXT: s_lshl_b32 s7, s40, 16
; SI-NEXT: s_or_b32 s57, s5, s7
; SI-NEXT: s_and_b32 s5, s21, 0xffff
; SI-NEXT: s_lshl_b32 s7, s41, 16
; SI-NEXT: s_or_b32 s59, s5, s7
; SI-NEXT: s_and_b32 s5, s23, 0xffff
; SI-NEXT: s_lshl_b32 s7, s42, 16
; SI-NEXT: s_or_b32 s61, s5, s7
; SI-NEXT: s_and_b32 s5, s25, 0xffff
; SI-NEXT: s_lshl_b32 s7, s43, 16
; SI-NEXT: s_or_b32 s63, s5, s7
; SI-NEXT: s_and_b32 s5, s27, 0xffff
; SI-NEXT: s_lshl_b32 s7, s44, 16
; SI-NEXT: s_or_b32 s73, s5, s7
; SI-NEXT: s_lshr_b64 s[16:17], s[46:47], 16
; SI-NEXT: s_lshr_b64 s[18:19], s[56:57], 16
; SI-NEXT: s_lshr_b64 s[20:21], s[58:59], 16
; SI-NEXT: s_lshr_b64 s[22:23], s[60:61], 16
; SI-NEXT: s_lshr_b64 s[24:25], s[62:63], 16
; SI-NEXT: s_lshr_b64 s[26:27], s[72:73], 16
; SI-NEXT: s_mov_b32 s15, s47
; SI-NEXT: s_mov_b32 s13, s57
; SI-NEXT: s_mov_b32 s11, s59
; SI-NEXT: s_mov_b32 s9, s61
; SI-NEXT: s_mov_b32 s7, s63
; SI-NEXT: s_mov_b32 s5, s73
; SI-NEXT: s_branch .LBB57_3
; SI-NEXT: .LBB57_2: ; %cmp.true
; SI-NEXT: s_add_i32 s26, s26, 3
; SI-NEXT: s_and_b32 s5, s26, 0xffff
; SI-NEXT: s_lshl_b32 s4, s4, 16
; SI-NEXT: s_add_i32 s27, s27, 3
; SI-NEXT: s_or_b32 s4, s4, s5
; SI-NEXT: s_and_b32 s5, s27, 0xffff
; SI-NEXT: s_lshl_b32 s7, s44, 16
; SI-NEXT: s_add_i32 s24, s24, 3
; SI-NEXT: s_or_b32 s5, s7, s5
; SI-NEXT: s_and_b32 s7, s24, 0xffff
; SI-NEXT: s_lshl_b32 s6, s6, 16
; SI-NEXT: s_add_i32 s25, s25, 3
; SI-NEXT: s_or_b32 s6, s6, s7
; SI-NEXT: s_and_b32 s7, s25, 0xffff
; SI-NEXT: s_lshl_b32 s9, s43, 16
; SI-NEXT: s_add_i32 s22, s22, 3
; SI-NEXT: s_or_b32 s7, s9, s7
; SI-NEXT: s_and_b32 s9, s22, 0xffff
; SI-NEXT: s_lshl_b32 s8, s8, 16
; SI-NEXT: s_add_i32 s23, s23, 3
; SI-NEXT: s_or_b32 s8, s8, s9
; SI-NEXT: s_and_b32 s9, s23, 0xffff
; SI-NEXT: s_lshl_b32 s11, s42, 16
; SI-NEXT: s_add_i32 s20, s20, 3
; SI-NEXT: s_or_b32 s9, s11, s9
; SI-NEXT: s_and_b32 s11, s20, 0xffff
; SI-NEXT: s_lshl_b32 s10, s10, 16
; SI-NEXT: s_add_i32 s21, s21, 3
; SI-NEXT: s_or_b32 s10, s10, s11
; SI-NEXT: s_and_b32 s11, s21, 0xffff
; SI-NEXT: s_lshl_b32 s13, s41, 16
; SI-NEXT: s_add_i32 s18, s18, 3
; SI-NEXT: s_or_b32 s11, s13, s11
; SI-NEXT: s_and_b32 s13, s18, 0xffff
; SI-NEXT: s_lshl_b32 s12, s12, 16
; SI-NEXT: s_add_i32 s19, s19, 3
; SI-NEXT: s_or_b32 s12, s12, s13
; SI-NEXT: s_and_b32 s13, s19, 0xffff
; SI-NEXT: s_lshl_b32 s15, s40, 16
; SI-NEXT: s_add_i32 s16, s16, 3
; SI-NEXT: s_or_b32 s13, s15, s13
; SI-NEXT: s_and_b32 s15, s16, 0xffff
; SI-NEXT: s_lshl_b32 s14, s14, 16
; SI-NEXT: s_add_i32 s17, s17, 3
; SI-NEXT: s_or_b32 s14, s14, s15
; SI-NEXT: s_and_b32 s15, s17, 0xffff
; SI-NEXT: s_lshl_b32 s16, s29, 16
; SI-NEXT: s_or_b32 s15, s16, s15
; SI-NEXT: s_add_i32 s4, s4, 0x30000
; SI-NEXT: s_add_i32 s5, s5, 0x30000
; SI-NEXT: s_add_i32 s6, s6, 0x30000
; SI-NEXT: s_add_i32 s7, s7, 0x30000
; SI-NEXT: s_add_i32 s8, s8, 0x30000
; SI-NEXT: s_add_i32 s9, s9, 0x30000
; SI-NEXT: s_add_i32 s10, s10, 0x30000
; SI-NEXT: s_add_i32 s11, s11, 0x30000
; SI-NEXT: s_add_i32 s12, s12, 0x30000
; SI-NEXT: s_add_i32 s13, s13, 0x30000
; SI-NEXT: s_add_i32 s14, s14, 0x30000
; SI-NEXT: s_add_i32 s15, s15, 0x30000
; SI-NEXT: s_lshr_b64 s[16:17], s[14:15], 16
; SI-NEXT: s_lshr_b64 s[18:19], s[12:13], 16
; SI-NEXT: s_lshr_b64 s[20:21], s[10:11], 16
; SI-NEXT: s_lshr_b64 s[22:23], s[8:9], 16
; SI-NEXT: s_lshr_b64 s[24:25], s[6:7], 16
; SI-NEXT: s_lshr_b64 s[26:27], s[4:5], 16
; SI-NEXT: s_lshr_b32 s29, s15, 16
; SI-NEXT: s_lshr_b32 s40, s13, 16
; SI-NEXT: s_lshr_b32 s41, s11, 16
; SI-NEXT: s_lshr_b32 s42, s9, 16
; SI-NEXT: s_lshr_b32 s43, s7, 16
; SI-NEXT: s_lshr_b32 s44, s5, 16
; SI-NEXT: .LBB57_3: ; %end
; SI-NEXT: s_and_b32 s14, s14, 0xffff
; SI-NEXT: s_lshl_b32 s16, s16, 16
; SI-NEXT: s_or_b32 s14, s14, s16
; SI-NEXT: s_and_b32 s15, s15, 0xffff
; SI-NEXT: s_lshl_b32 s16, s29, 16
; SI-NEXT: s_or_b32 s15, s15, s16
; SI-NEXT: s_and_b32 s12, s12, 0xffff
; SI-NEXT: s_lshl_b32 s16, s18, 16
; SI-NEXT: s_or_b32 s12, s12, s16
; SI-NEXT: s_and_b32 s13, s13, 0xffff
; SI-NEXT: s_lshl_b32 s16, s40, 16
; SI-NEXT: s_or_b32 s13, s13, s16
; SI-NEXT: s_and_b32 s10, s10, 0xffff
; SI-NEXT: s_lshl_b32 s16, s20, 16
; SI-NEXT: s_or_b32 s10, s10, s16
; SI-NEXT: s_and_b32 s11, s11, 0xffff
; SI-NEXT: s_lshl_b32 s16, s41, 16
; SI-NEXT: s_or_b32 s11, s11, s16
; SI-NEXT: s_and_b32 s8, s8, 0xffff
; SI-NEXT: s_lshl_b32 s16, s22, 16
; SI-NEXT: s_or_b32 s8, s8, s16
; SI-NEXT: s_and_b32 s9, s9, 0xffff
; SI-NEXT: s_lshl_b32 s16, s42, 16
; SI-NEXT: s_or_b32 s9, s9, s16
; SI-NEXT: s_and_b32 s6, s6, 0xffff
; SI-NEXT: s_lshl_b32 s16, s24, 16
; SI-NEXT: s_or_b32 s6, s6, s16
; SI-NEXT: s_and_b32 s7, s7, 0xffff
; SI-NEXT: s_lshl_b32 s16, s43, 16
; SI-NEXT: s_or_b32 s7, s7, s16
; SI-NEXT: s_and_b32 s4, s4, 0xffff
; SI-NEXT: s_lshl_b32 s16, s26, 16
; SI-NEXT: s_or_b32 s4, s4, s16
; SI-NEXT: s_and_b32 s5, s5, 0xffff
; SI-NEXT: s_lshl_b32 s16, s44, 16
; SI-NEXT: s_or_b32 s5, s5, s16
; SI-NEXT: v_mov_b32_e32 v0, s14
; SI-NEXT: v_mov_b32_e32 v1, s15
; SI-NEXT: v_mov_b32_e32 v2, s12
; SI-NEXT: v_mov_b32_e32 v3, s13
; SI-NEXT: v_mov_b32_e32 v4, s10
; SI-NEXT: v_mov_b32_e32 v5, s11
; SI-NEXT: v_mov_b32_e32 v6, s8
; SI-NEXT: v_mov_b32_e32 v7, s9
; SI-NEXT: v_mov_b32_e32 v8, s6
; SI-NEXT: v_mov_b32_e32 v9, s7
; SI-NEXT: v_mov_b32_e32 v10, s4
; SI-NEXT: v_mov_b32_e32 v11, s5
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24i16_to_v24f16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc1 .LBB57_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: s_and_b32 s4, s16, 0xffff0000
; VI-NEXT: s_add_i32 s5, s16, 3
; VI-NEXT: s_and_b32 s6, s17, 0xffff0000
; VI-NEXT: s_add_i32 s7, s17, 3
; VI-NEXT: s_and_b32 s8, s18, 0xffff0000
; VI-NEXT: s_add_i32 s9, s18, 3
; VI-NEXT: s_and_b32 s10, s19, 0xffff0000
; VI-NEXT: s_add_i32 s11, s19, 3
; VI-NEXT: s_and_b32 s12, s20, 0xffff0000
; VI-NEXT: s_add_i32 s13, s20, 3
; VI-NEXT: s_and_b32 s14, s21, 0xffff0000
; VI-NEXT: s_add_i32 s15, s21, 3
; VI-NEXT: s_and_b32 s16, s22, 0xffff0000
; VI-NEXT: s_add_i32 s17, s22, 3
; VI-NEXT: s_and_b32 s18, s23, 0xffff0000
; VI-NEXT: s_add_i32 s19, s23, 3
; VI-NEXT: s_and_b32 s20, s24, 0xffff0000
; VI-NEXT: s_add_i32 s21, s24, 3
; VI-NEXT: s_and_b32 s22, s25, 0xffff0000
; VI-NEXT: s_add_i32 s23, s25, 3
; VI-NEXT: s_and_b32 s24, s26, 0xffff0000
; VI-NEXT: s_add_i32 s25, s26, 3
; VI-NEXT: s_and_b32 s26, s27, 0xffff0000
; VI-NEXT: s_add_i32 s27, s27, 3
; VI-NEXT: s_and_b32 s27, s27, 0xffff
; VI-NEXT: s_and_b32 s25, s25, 0xffff
; VI-NEXT: s_and_b32 s23, s23, 0xffff
; VI-NEXT: s_and_b32 s21, s21, 0xffff
; VI-NEXT: s_and_b32 s19, s19, 0xffff
; VI-NEXT: s_and_b32 s17, s17, 0xffff
; VI-NEXT: s_and_b32 s15, s15, 0xffff
; VI-NEXT: s_and_b32 s13, s13, 0xffff
; VI-NEXT: s_and_b32 s11, s11, 0xffff
; VI-NEXT: s_and_b32 s9, s9, 0xffff
; VI-NEXT: s_and_b32 s7, s7, 0xffff
; VI-NEXT: s_and_b32 s5, s5, 0xffff
; VI-NEXT: s_or_b32 s26, s26, s27
; VI-NEXT: s_or_b32 s24, s24, s25
; VI-NEXT: s_or_b32 s22, s22, s23
; VI-NEXT: s_or_b32 s20, s20, s21
; VI-NEXT: s_or_b32 s18, s18, s19
; VI-NEXT: s_or_b32 s16, s16, s17
; VI-NEXT: s_or_b32 s14, s14, s15
; VI-NEXT: s_or_b32 s12, s12, s13
; VI-NEXT: s_or_b32 s10, s10, s11
; VI-NEXT: s_or_b32 s8, s8, s9
; VI-NEXT: s_or_b32 s6, s6, s7
; VI-NEXT: s_or_b32 s4, s4, s5
; VI-NEXT: s_add_i32 s27, s26, 0x30000
; VI-NEXT: s_add_i32 s26, s24, 0x30000
; VI-NEXT: s_add_i32 s25, s22, 0x30000
; VI-NEXT: s_add_i32 s24, s20, 0x30000
; VI-NEXT: s_add_i32 s23, s18, 0x30000
; VI-NEXT: s_add_i32 s22, s16, 0x30000
; VI-NEXT: s_add_i32 s21, s14, 0x30000
; VI-NEXT: s_add_i32 s20, s12, 0x30000
; VI-NEXT: s_add_i32 s19, s10, 0x30000
; VI-NEXT: s_add_i32 s18, s8, 0x30000
; VI-NEXT: s_add_i32 s17, s6, 0x30000
; VI-NEXT: s_add_i32 s16, s4, 0x30000
; VI-NEXT: .LBB57_2: ; %end
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24i16_to_v24f16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB57_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB57_2: ; %cmp.true
; GFX9-NEXT: v_pk_add_u16 v11, s27, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v10, s26, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v9, s25, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v8, s24, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v7, s23, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v6, s22, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v5, s21, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v4, s20, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v3, s19, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v2, s18, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v1, s17, 3 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_u16 v0, s16, 3 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24i16_to_v24f16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB57_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB57_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_u16 v11, s23, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v10, s22, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v9, s21, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v8, s20, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v7, s19, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v6, s18, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v5, s17, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v4, s16, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v3, s15, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v2, s14, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v1, s13, 3 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_u16 v0, s12, 3 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = add <24 x i16> %a, splat (i16 3)
%a2 = bitcast <24 x i16> %a1 to <24 x half>
br label %end
cmp.false:
%a3 = bitcast <24 x i16> %a to <24 x half>
br label %end
end:
%phi = phi <24 x half> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x half> %phi
}
define <24 x i16> @bitcast_v24f16_to_v24i16(<24 x half> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v24i16:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: v_lshrrev_b32_e32 v13, 16, v11
; SI-NEXT: v_lshrrev_b32_e32 v15, 16, v10
; SI-NEXT: v_lshrrev_b32_e32 v14, 16, v9
; SI-NEXT: v_lshrrev_b32_e32 v18, 16, v8
; SI-NEXT: v_lshrrev_b32_e32 v16, 16, v7
; SI-NEXT: v_lshrrev_b32_e32 v19, 16, v6
; SI-NEXT: v_lshrrev_b32_e32 v17, 16, v5
; SI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; SI-NEXT: v_lshrrev_b32_e32 v21, 16, v4
; SI-NEXT: v_lshrrev_b32_e32 v12, 16, v3
; SI-NEXT: v_lshrrev_b32_e32 v22, 16, v2
; SI-NEXT: v_lshrrev_b32_e32 v20, 16, v1
; SI-NEXT: v_lshrrev_b32_e32 v23, 16, v0
; SI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; SI-NEXT: s_cbranch_execz .LBB58_2
; SI-NEXT: ; %bb.1: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v13, v13
; SI-NEXT: v_cvt_f32_f16_e32 v11, v11
; SI-NEXT: v_cvt_f32_f16_e32 v14, v14
; SI-NEXT: v_cvt_f32_f16_e32 v9, v9
; SI-NEXT: v_cvt_f32_f16_e32 v16, v16
; SI-NEXT: v_cvt_f32_f16_e32 v7, v7
; SI-NEXT: v_cvt_f32_f16_e32 v17, v17
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_cvt_f32_f16_e32 v5, v5
; SI-NEXT: v_cvt_f32_f16_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_add_f32_e32 v14, 0x38000000, v14
; SI-NEXT: v_cvt_f32_f16_e32 v3, v3
; SI-NEXT: v_cvt_f32_f16_e32 v20, v20
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f16_f32_e32 v14, v14
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_add_f32_e32 v16, 0x38000000, v16
; SI-NEXT: v_cvt_f32_f16_e32 v1, v1
; SI-NEXT: v_cvt_f32_f16_e32 v23, v23
; SI-NEXT: v_cvt_f32_f16_e32 v22, v22
; SI-NEXT: v_cvt_f32_f16_e32 v21, v21
; SI-NEXT: v_cvt_f32_f16_e32 v19, v19
; SI-NEXT: v_cvt_f32_f16_e32 v18, v18
; SI-NEXT: v_cvt_f32_f16_e32 v15, v15
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_cvt_f16_f32_e32 v16, v16
; SI-NEXT: v_add_f32_e32 v7, 0x38000000, v7
; SI-NEXT: v_add_f32_e32 v17, 0x38000000, v17
; SI-NEXT: v_cvt_f32_f16_e32 v0, v0
; SI-NEXT: v_cvt_f32_f16_e32 v2, v2
; SI-NEXT: v_cvt_f32_f16_e32 v4, v4
; SI-NEXT: v_cvt_f32_f16_e32 v6, v6
; SI-NEXT: v_cvt_f32_f16_e32 v8, v8
; SI-NEXT: v_cvt_f32_f16_e32 v10, v10
; SI-NEXT: v_cvt_f16_f32_e32 v7, v7
; SI-NEXT: v_cvt_f16_f32_e32 v17, v17
; SI-NEXT: v_add_f32_e32 v5, 0x38000000, v5
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_lshlrev_b32_e32 v24, 16, v13
; SI-NEXT: v_cvt_f16_f32_e32 v5, v5
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_add_f32_e32 v20, 0x38000000, v20
; SI-NEXT: v_or_b32_e32 v11, v11, v24
; SI-NEXT: v_lshlrev_b32_e32 v24, 16, v14
; SI-NEXT: v_cvt_f16_f32_e32 v3, v3
; SI-NEXT: v_cvt_f16_f32_e32 v20, v20
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_add_f32_e32 v23, 0x38000000, v23
; SI-NEXT: v_add_f32_e32 v22, 0x38000000, v22
; SI-NEXT: v_add_f32_e32 v21, 0x38000000, v21
; SI-NEXT: v_add_f32_e32 v19, 0x38000000, v19
; SI-NEXT: v_add_f32_e32 v18, 0x38000000, v18
; SI-NEXT: v_add_f32_e32 v15, 0x38000000, v15
; SI-NEXT: v_or_b32_e32 v9, v9, v24
; SI-NEXT: v_lshlrev_b32_e32 v24, 16, v16
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_cvt_f16_f32_e32 v23, v23
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_cvt_f16_f32_e32 v22, v22
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v21, v21
; SI-NEXT: v_add_f32_e32 v4, 0x38000000, v4
; SI-NEXT: v_cvt_f16_f32_e32 v19, v19
; SI-NEXT: v_add_f32_e32 v6, 0x38000000, v6
; SI-NEXT: v_cvt_f16_f32_e32 v18, v18
; SI-NEXT: v_add_f32_e32 v8, 0x38000000, v8
; SI-NEXT: v_cvt_f16_f32_e32 v15, v15
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_or_b32_e32 v7, v7, v24
; SI-NEXT: v_lshlrev_b32_e32 v24, 16, v17
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_cvt_f16_f32_e32 v2, v2
; SI-NEXT: v_cvt_f16_f32_e32 v4, v4
; SI-NEXT: v_cvt_f16_f32_e32 v6, v6
; SI-NEXT: v_cvt_f16_f32_e32 v8, v8
; SI-NEXT: v_cvt_f16_f32_e32 v10, v10
; SI-NEXT: v_or_b32_e32 v5, v5, v24
; SI-NEXT: v_lshlrev_b32_e32 v24, 16, v12
; SI-NEXT: v_or_b32_e32 v3, v3, v24
; SI-NEXT: v_lshlrev_b32_e32 v24, 16, v20
; SI-NEXT: v_or_b32_e32 v1, v1, v24
; SI-NEXT: v_lshlrev_b32_e32 v23, 16, v23
; SI-NEXT: v_lshlrev_b32_e32 v22, 16, v22
; SI-NEXT: v_lshlrev_b32_e32 v21, 16, v21
; SI-NEXT: v_lshlrev_b32_e32 v19, 16, v19
; SI-NEXT: v_lshlrev_b32_e32 v18, 16, v18
; SI-NEXT: v_lshlrev_b32_e32 v15, 16, v15
; SI-NEXT: v_or_b32_e32 v0, v0, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v22
; SI-NEXT: v_or_b32_e32 v4, v4, v21
; SI-NEXT: v_or_b32_e32 v6, v6, v19
; SI-NEXT: v_or_b32_e32 v8, v8, v18
; SI-NEXT: v_or_b32_e32 v10, v10, v15
; SI-NEXT: v_alignbit_b32 v23, v1, v23, 16
; SI-NEXT: v_alignbit_b32 v22, v3, v22, 16
; SI-NEXT: v_alignbit_b32 v21, v5, v21, 16
; SI-NEXT: v_alignbit_b32 v19, v7, v19, 16
; SI-NEXT: v_alignbit_b32 v18, v9, v18, 16
; SI-NEXT: v_alignbit_b32 v15, v11, v15, 16
; SI-NEXT: .LBB58_2: ; %end
; SI-NEXT: s_or_b64 exec, exec, s[4:5]
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v3, v3, v12
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v4
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v21
; SI-NEXT: v_or_b32_e32 v4, v4, v12
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v17
; SI-NEXT: v_or_b32_e32 v5, v5, v12
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v6
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v19
; SI-NEXT: v_or_b32_e32 v6, v6, v12
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v16
; SI-NEXT: v_or_b32_e32 v7, v7, v12
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v18
; SI-NEXT: v_or_b32_e32 v8, v8, v12
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v14
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
; SI-NEXT: v_or_b32_e32 v9, v9, v12
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v10
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v15
; SI-NEXT: v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT: v_lshlrev_b32_e32 v23, 16, v23
; SI-NEXT: v_or_b32_e32 v1, v1, v20
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT: v_lshlrev_b32_e32 v20, 16, v22
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v13
; SI-NEXT: v_or_b32_e32 v0, v0, v23
; SI-NEXT: v_or_b32_e32 v2, v2, v20
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v24i16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; VI-NEXT: s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; VI-NEXT: s_cbranch_execz .LBB58_2
; VI-NEXT: ; %bb.1: ; %cmp.true
; VI-NEXT: v_mov_b32_e32 v13, 0x200
; VI-NEXT: v_add_f16_e32 v19, 0x200, v11
; VI-NEXT: v_add_f16_sdwa v11, v11, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v11, v19, v11
; VI-NEXT: v_add_f16_e32 v19, 0x200, v10
; VI-NEXT: v_add_f16_sdwa v10, v10, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v10, v19, v10
; VI-NEXT: v_add_f16_e32 v19, 0x200, v9
; VI-NEXT: v_add_f16_sdwa v9, v9, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v9, v19, v9
; VI-NEXT: v_add_f16_e32 v19, 0x200, v8
; VI-NEXT: v_add_f16_sdwa v8, v8, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v12, 0x200, v0
; VI-NEXT: v_add_f16_sdwa v0, v0, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v14, 0x200, v1
; VI-NEXT: v_add_f16_sdwa v1, v1, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v15, 0x200, v2
; VI-NEXT: v_add_f16_sdwa v2, v2, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v16, 0x200, v3
; VI-NEXT: v_add_f16_sdwa v3, v3, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v17, 0x200, v4
; VI-NEXT: v_add_f16_sdwa v4, v4, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v18, 0x200, v5
; VI-NEXT: v_add_f16_sdwa v5, v5, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v8, v19, v8
; VI-NEXT: v_add_f16_e32 v19, 0x200, v6
; VI-NEXT: v_add_f16_sdwa v6, v6, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_sdwa v13, v7, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v7, 0x200, v7
; VI-NEXT: v_or_b32_e32 v7, v7, v13
; VI-NEXT: v_or_b32_e32 v6, v19, v6
; VI-NEXT: v_or_b32_e32 v5, v18, v5
; VI-NEXT: v_or_b32_e32 v4, v17, v4
; VI-NEXT: v_or_b32_e32 v3, v16, v3
; VI-NEXT: v_or_b32_e32 v2, v15, v2
; VI-NEXT: v_or_b32_e32 v1, v14, v1
; VI-NEXT: v_or_b32_e32 v0, v12, v0
; VI-NEXT: .LBB58_2: ; %end
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v24i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_ne_u32_e32 vcc, 0, v12
; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT: s_andn2_saveexec_b64 s[4:5], s[4:5]
; GFX9-NEXT: s_cbranch_execz .LBB58_2
; GFX9-NEXT: ; %bb.1: ; %cmp.true
; GFX9-NEXT: s_movk_i32 s6, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, v11, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, v10, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, v9, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, v8, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, v7, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, v6, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, v5, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, v4, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, v3, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, v2, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, v1, s6 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, v0, s6 op_sel_hi:[1,0]
; GFX9-NEXT: .LBB58_2: ; %end
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v24i16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s0, exec_lo
; GFX11-NEXT: v_cmpx_ne_u32_e32 0, v12
; GFX11-NEXT: s_xor_b32 s0, exec_lo, s0
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_not1_saveexec_b32 s0, s0
; GFX11-NEXT: s_cbranch_execz .LBB58_2
; GFX11-NEXT: ; %bb.1: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, v11, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, v10, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, v9, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, v8, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, v7, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, v6, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, v5, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, v4, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, v3, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, v2, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, v1, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, v0, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: .LBB58_2: ; %end
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
define inreg <24 x i16> @bitcast_v24f16_to_v24i16_scalar(<24 x half> inreg %a, i32 inreg %b) #0 {
; SI-LABEL: bitcast_v24f16_to_v24i16_scalar:
; SI: ; %bb.0:
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s11, s27, 16
; SI-NEXT: s_lshr_b32 s6, s26, 16
; SI-NEXT: s_lshr_b32 s8, s25, 16
; SI-NEXT: s_lshr_b32 s7, s24, 16
; SI-NEXT: s_lshr_b32 s10, s23, 16
; SI-NEXT: s_lshr_b32 s12, s22, 16
; SI-NEXT: s_lshr_b32 s9, s21, 16
; SI-NEXT: s_lshr_b32 s13, s20, 16
; SI-NEXT: s_lshr_b32 s5, s19, 16
; SI-NEXT: s_lshr_b32 s14, s18, 16
; SI-NEXT: s_lshr_b32 s4, s17, 16
; SI-NEXT: s_lshr_b32 s15, s16, 16
; SI-NEXT: s_cmp_lg_u32 s28, 0
; SI-NEXT: s_cbranch_scc0 .LBB59_2
; SI-NEXT: ; %bb.1: ; %cmp.false
; SI-NEXT: v_mov_b32_e32 v24, s11
; SI-NEXT: v_mov_b32_e32 v28, s8
; SI-NEXT: v_mov_b32_e32 v25, s10
; SI-NEXT: v_mov_b32_e32 v26, s9
; SI-NEXT: v_mov_b32_e32 v27, s5
; SI-NEXT: v_mov_b32_e32 v29, s4
; SI-NEXT: v_mov_b32_e32 v1, s17
; SI-NEXT: v_mov_b32_e32 v3, s19
; SI-NEXT: v_mov_b32_e32 v5, s21
; SI-NEXT: v_mov_b32_e32 v7, s23
; SI-NEXT: v_mov_b32_e32 v9, s25
; SI-NEXT: v_mov_b32_e32 v11, s27
; SI-NEXT: v_mov_b32_e32 v30, s26
; SI-NEXT: v_mov_b32_e32 v31, s24
; SI-NEXT: v_mov_b32_e32 v33, s22
; SI-NEXT: v_mov_b32_e32 v35, s20
; SI-NEXT: v_mov_b32_e32 v32, s18
; SI-NEXT: v_mov_b32_e32 v34, s16
; SI-NEXT: v_mov_b32_e32 v22, s15
; SI-NEXT: v_mov_b32_e32 v20, s14
; SI-NEXT: v_mov_b32_e32 v18, s13
; SI-NEXT: v_mov_b32_e32 v16, s12
; SI-NEXT: v_mov_b32_e32 v14, s7
; SI-NEXT: v_mov_b32_e32 v12, s6
; SI-NEXT: s_branch .LBB59_3
; SI-NEXT: .LBB59_2: ; %cmp.true
; SI-NEXT: v_cvt_f32_f16_e32 v1, s14
; SI-NEXT: v_cvt_f32_f16_e32 v2, s13
; SI-NEXT: v_cvt_f32_f16_e32 v3, s12
; SI-NEXT: v_cvt_f32_f16_e32 v5, s23
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v2
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_cvt_f16_f32_e32 v4, v2
; SI-NEXT: v_add_f32_e32 v2, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v3, v2
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v1
; SI-NEXT: v_cvt_f32_f16_e32 v1, s7
; SI-NEXT: v_cvt_f32_f16_e32 v11, s19
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v3
; SI-NEXT: v_cvt_f32_f16_e32 v3, s6
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v8, v1
; SI-NEXT: v_cvt_f32_f16_e32 v1, s11
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v12, v3
; SI-NEXT: v_cvt_f32_f16_e32 v3, s10
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v24, v1
; SI-NEXT: v_cvt_f32_f16_e32 v13, s17
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v25, v1
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v5
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_cvt_f32_f16_e32 v3, s9
; SI-NEXT: v_lshlrev_b32_e32 v5, 16, v25
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v11
; SI-NEXT: v_or_b32_e32 v7, v1, v5
; SI-NEXT: v_cvt_f32_f16_e32 v1, s21
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f32_f16_e32 v5, s5
; SI-NEXT: v_cvt_f16_f32_e32 v26, v3
; SI-NEXT: v_add_f32_e32 v1, 0x38000000, v1
; SI-NEXT: v_cvt_f16_f32_e32 v1, v1
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v5
; SI-NEXT: v_cvt_f16_f32_e32 v27, v3
; SI-NEXT: v_lshlrev_b32_e32 v3, 16, v26
; SI-NEXT: v_or_b32_e32 v5, v1, v3
; SI-NEXT: v_cvt_f32_f16_e32 v3, s4
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f32_f16_e32 v9, s8
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v27
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v3
; SI-NEXT: v_cvt_f16_f32_e32 v29, v3
; SI-NEXT: v_add_f32_e32 v3, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v3
; SI-NEXT: v_or_b32_e32 v3, v11, v1
; SI-NEXT: v_cvt_f32_f16_e32 v11, s25
; SI-NEXT: v_lshlrev_b32_e32 v1, 16, v29
; SI-NEXT: v_or_b32_e32 v1, v13, v1
; SI-NEXT: v_cvt_f32_f16_e32 v13, s27
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v9
; SI-NEXT: v_cvt_f16_f32_e32 v28, v9
; SI-NEXT: v_add_f32_e32 v9, 0x38000000, v11
; SI-NEXT: v_cvt_f32_f16_e32 v0, s15
; SI-NEXT: v_cvt_f32_f16_e32 v10, s16
; SI-NEXT: v_cvt_f16_f32_e32 v9, v9
; SI-NEXT: v_add_f32_e32 v11, 0x38000000, v13
; SI-NEXT: v_cvt_f16_f32_e32 v11, v11
; SI-NEXT: v_cvt_f32_f16_e32 v14, s18
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v28
; SI-NEXT: v_add_f32_e32 v0, 0x38000000, v0
; SI-NEXT: v_or_b32_e32 v9, v9, v13
; SI-NEXT: v_lshlrev_b32_e32 v13, 16, v24
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v10
; SI-NEXT: v_cvt_f16_f32_e32 v0, v0
; SI-NEXT: v_or_b32_e32 v11, v11, v13
; SI-NEXT: v_cvt_f16_f32_e32 v13, v10
; SI-NEXT: v_add_f32_e32 v10, 0x38000000, v14
; SI-NEXT: v_cvt_f16_f32_e32 v14, v10
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v12
; SI-NEXT: v_cvt_f32_f16_e32 v12, s20
; SI-NEXT: v_or_b32_e32 v34, v13, v0
; SI-NEXT: v_or_b32_e32 v32, v14, v2
; SI-NEXT: v_cvt_f32_f16_e32 v13, s22
; SI-NEXT: v_cvt_f32_f16_e32 v14, s24
; SI-NEXT: v_cvt_f32_f16_e32 v15, s26
; SI-NEXT: v_add_f32_e32 v12, 0x38000000, v12
; SI-NEXT: v_add_f32_e32 v13, 0x38000000, v13
; SI-NEXT: v_add_f32_e32 v14, 0x38000000, v14
; SI-NEXT: v_add_f32_e32 v15, 0x38000000, v15
; SI-NEXT: v_cvt_f16_f32_e32 v12, v12
; SI-NEXT: v_cvt_f16_f32_e32 v13, v13
; SI-NEXT: v_cvt_f16_f32_e32 v14, v14
; SI-NEXT: v_cvt_f16_f32_e32 v15, v15
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v4
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v8
; SI-NEXT: v_or_b32_e32 v35, v12, v4
; SI-NEXT: v_or_b32_e32 v33, v13, v6
; SI-NEXT: v_or_b32_e32 v31, v14, v8
; SI-NEXT: v_or_b32_e32 v30, v15, v10
; SI-NEXT: v_lshr_b64 v[22:23], v[0:1], 16
; SI-NEXT: v_lshr_b64 v[20:21], v[2:3], 16
; SI-NEXT: v_lshr_b64 v[18:19], v[4:5], 16
; SI-NEXT: v_lshr_b64 v[16:17], v[6:7], 16
; SI-NEXT: v_lshr_b64 v[14:15], v[8:9], 16
; SI-NEXT: v_lshr_b64 v[12:13], v[10:11], 16
; SI-NEXT: .LBB59_3: ; %end
; SI-NEXT: v_lshlrev_b32_e32 v0, 16, v22
; SI-NEXT: v_and_b32_e32 v2, 0xffff, v34
; SI-NEXT: v_or_b32_e32 v0, v2, v0
; SI-NEXT: v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v29
; SI-NEXT: v_or_b32_e32 v1, v1, v2
; SI-NEXT: v_lshlrev_b32_e32 v2, 16, v20
; SI-NEXT: v_and_b32_e32 v4, 0xffff, v32
; SI-NEXT: v_or_b32_e32 v2, v4, v2
; SI-NEXT: v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v27
; SI-NEXT: v_or_b32_e32 v3, v3, v4
; SI-NEXT: v_lshlrev_b32_e32 v4, 16, v18
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v35
; SI-NEXT: v_or_b32_e32 v4, v6, v4
; SI-NEXT: v_and_b32_e32 v5, 0xffff, v5
; SI-NEXT: v_lshlrev_b32_e32 v6, 16, v26
; SI-NEXT: v_or_b32_e32 v5, v5, v6
; SI-NEXT: v_and_b32_e32 v6, 0xffff, v33
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v16
; SI-NEXT: v_or_b32_e32 v6, v6, v8
; SI-NEXT: v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT: v_lshlrev_b32_e32 v8, 16, v25
; SI-NEXT: v_or_b32_e32 v7, v7, v8
; SI-NEXT: v_and_b32_e32 v8, 0xffff, v31
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v14
; SI-NEXT: v_or_b32_e32 v8, v8, v10
; SI-NEXT: v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT: v_lshlrev_b32_e32 v10, 16, v28
; SI-NEXT: v_or_b32_e32 v9, v9, v10
; SI-NEXT: v_and_b32_e32 v10, 0xffff, v30
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT: v_or_b32_e32 v10, v10, v12
; SI-NEXT: v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT: v_lshlrev_b32_e32 v12, 16, v24
; SI-NEXT: v_or_b32_e32 v11, v11, v12
; SI-NEXT: s_setpc_b64 s[30:31]
;
; VI-LABEL: bitcast_v24f16_to_v24i16_scalar:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: s_cmp_lg_u32 s28, 0
; VI-NEXT: s_cbranch_scc0 .LBB59_2
; VI-NEXT: ; %bb.1: ; %cmp.false
; VI-NEXT: v_mov_b32_e32 v0, s16
; VI-NEXT: v_mov_b32_e32 v1, s17
; VI-NEXT: v_mov_b32_e32 v2, s18
; VI-NEXT: v_mov_b32_e32 v3, s19
; VI-NEXT: v_mov_b32_e32 v4, s20
; VI-NEXT: v_mov_b32_e32 v5, s21
; VI-NEXT: v_mov_b32_e32 v6, s22
; VI-NEXT: v_mov_b32_e32 v7, s23
; VI-NEXT: v_mov_b32_e32 v8, s24
; VI-NEXT: v_mov_b32_e32 v9, s25
; VI-NEXT: v_mov_b32_e32 v10, s26
; VI-NEXT: v_mov_b32_e32 v11, s27
; VI-NEXT: v_mov_b32_e32 v12, s28
; VI-NEXT: v_mov_b32_e32 v13, s29
; VI-NEXT: v_mov_b32_e32 v14, s30
; VI-NEXT: v_mov_b32_e32 v15, s31
; VI-NEXT: s_setpc_b64 s[30:31]
; VI-NEXT: .LBB59_2: ; %cmp.true
; VI-NEXT: s_lshr_b32 s5, s26, 16
; VI-NEXT: v_mov_b32_e32 v0, 0x200
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: s_lshr_b32 s5, s27, 16
; VI-NEXT: v_add_f16_e32 v1, s26, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_mov_b32_e32 v4, s5
; VI-NEXT: s_lshr_b32 s5, s25, 16
; VI-NEXT: v_or_b32_e32 v10, v1, v2
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: v_add_f16_e32 v1, s25, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: s_lshr_b32 s5, s24, 16
; VI-NEXT: v_or_b32_e32 v9, v1, v2
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: v_add_f16_e32 v1, s24, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: s_lshr_b32 s5, s23, 16
; VI-NEXT: v_or_b32_e32 v8, v1, v2
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: v_add_f16_e32 v1, s23, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: s_lshr_b32 s5, s22, 16
; VI-NEXT: v_or_b32_e32 v7, v1, v2
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: v_add_f16_e32 v1, s22, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: s_lshr_b32 s5, s21, 16
; VI-NEXT: v_or_b32_e32 v6, v1, v2
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: v_add_f16_e32 v1, s21, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: s_lshr_b32 s5, s20, 16
; VI-NEXT: v_or_b32_e32 v5, v1, v2
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: v_add_f16_e32 v3, s27, v0
; VI-NEXT: v_add_f16_sdwa v4, v4, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v1, s20, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: s_lshr_b32 s5, s19, 16
; VI-NEXT: v_or_b32_e32 v11, v3, v4
; VI-NEXT: v_or_b32_e32 v4, v1, v2
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: v_add_f16_e32 v1, s19, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: s_lshr_b32 s5, s18, 16
; VI-NEXT: v_or_b32_e32 v3, v1, v2
; VI-NEXT: v_mov_b32_e32 v2, s5
; VI-NEXT: s_lshr_b32 s4, s16, 16
; VI-NEXT: v_add_f16_e32 v1, s18, v0
; VI-NEXT: v_add_f16_sdwa v2, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v2, v1, v2
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: s_lshr_b32 s4, s17, 16
; VI-NEXT: v_add_f16_sdwa v13, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_mov_b32_e32 v1, s4
; VI-NEXT: v_add_f16_e32 v12, s16, v0
; VI-NEXT: v_add_f16_sdwa v1, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_add_f16_e32 v0, s17, v0
; VI-NEXT: v_or_b32_e32 v1, v0, v1
; VI-NEXT: v_or_b32_e32 v0, v12, v13
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: bitcast_v24f16_to_v24i16_scalar:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_cmp_lg_u32 s28, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB59_2
; GFX9-NEXT: ; %bb.1: ; %cmp.false
; GFX9-NEXT: v_mov_b32_e32 v0, s16
; GFX9-NEXT: v_mov_b32_e32 v1, s17
; GFX9-NEXT: v_mov_b32_e32 v2, s18
; GFX9-NEXT: v_mov_b32_e32 v3, s19
; GFX9-NEXT: v_mov_b32_e32 v4, s20
; GFX9-NEXT: v_mov_b32_e32 v5, s21
; GFX9-NEXT: v_mov_b32_e32 v6, s22
; GFX9-NEXT: v_mov_b32_e32 v7, s23
; GFX9-NEXT: v_mov_b32_e32 v8, s24
; GFX9-NEXT: v_mov_b32_e32 v9, s25
; GFX9-NEXT: v_mov_b32_e32 v10, s26
; GFX9-NEXT: v_mov_b32_e32 v11, s27
; GFX9-NEXT: v_mov_b32_e32 v12, s28
; GFX9-NEXT: v_mov_b32_e32 v13, s29
; GFX9-NEXT: v_mov_b32_e32 v14, s30
; GFX9-NEXT: v_mov_b32_e32 v15, s31
; GFX9-NEXT: s_setpc_b64 s[30:31]
; GFX9-NEXT: .LBB59_2: ; %cmp.true
; GFX9-NEXT: v_mov_b32_e32 v0, 0x200
; GFX9-NEXT: v_pk_add_f16 v11, s27, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v10, s26, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v9, s25, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v8, s24, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v7, s23, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v6, s22, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v5, s21, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v4, s20, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v3, s19, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v2, s18, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v1, s17, v0 op_sel_hi:[1,0]
; GFX9-NEXT: v_pk_add_f16 v0, s16, v0 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: bitcast_v24f16_to_v24i16_scalar:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s15, s3
; GFX11-NEXT: s_mov_b32 s14, s2
; GFX11-NEXT: s_mov_b32 s13, s1
; GFX11-NEXT: s_mov_b32 s12, s0
; GFX11-NEXT: s_cmp_lg_u32 s24, 0
; GFX11-NEXT: s_cbranch_scc0 .LBB59_2
; GFX11-NEXT: ; %bb.1: ; %cmp.false
; GFX11-NEXT: v_dual_mov_b32 v0, s12 :: v_dual_mov_b32 v1, s13
; GFX11-NEXT: v_dual_mov_b32 v2, s14 :: v_dual_mov_b32 v3, s15
; GFX11-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
; GFX11-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
; GFX11-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
; GFX11-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
; GFX11-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
; GFX11-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
; GFX11-NEXT: s_setpc_b64 s[30:31]
; GFX11-NEXT: .LBB59_2: ; %cmp.true
; GFX11-NEXT: v_pk_add_f16 v11, s23, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v10, s22, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v9, s21, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v8, s20, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v7, s19, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v6, s18, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v5, s17, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v4, s16, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v3, s15, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v2, s14, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v1, s13, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: v_pk_add_f16 v0, s12, 0x200 op_sel_hi:[1,0]
; GFX11-NEXT: s_setpc_b64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
cmp.true:
%a1 = fadd <24 x half> %a, splat (half 0xH0200)
%a2 = bitcast <24 x half> %a1 to <24 x i16>
br label %end
cmp.false:
%a3 = bitcast <24 x half> %a to <24 x i16>
br label %end
end:
%phi = phi <24 x i16> [ %a2, %cmp.true ], [ %a3, %cmp.false ]
ret <24 x i16> %phi
}
attributes #0 = { nounwind }
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GFX11-FAKE16: {{.*}}
; GFX11-TRUE16: {{.*}}