blob: e10eb209cf32277747c23da30a17d441fb0721f1 [file] [edit]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc -mtriple=amdgpu9.50 < %s | FileCheck %s
define float @canonical_f32_dpp_i32(float %a, float %b) {
; CHECK-LABEL: canonical_f32_dpp_i32:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_add_f32_e32 v0, v0, v1
; CHECK-NEXT: s_nop 1
; CHECK-NEXT: v_mov_b32_dpp v0, v0 quad_perm:[1,0,3,2] row_mask:0xf bank_mask:0xf bound_ctrl:1
; CHECK-NEXT: s_setpc_b64 s[30:31]
%sum = fadd float %a, %b
%bits = bitcast float %sum to i32
%shuffle = call i32 @llvm.amdgcn.update.dpp.i32(
i32 poison, i32 %bits, i32 177, i32 15, i32 15, i1 true)
%value = bitcast i32 %shuffle to float
%result = call float @llvm.canonicalize.f32(float %value)
ret float %result
}
define float @canonical_f32_dpp_shift(float %a, float %b) {
; CHECK-LABEL: canonical_f32_dpp_shift:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_add_f32_e32 v0, v0, v1
; CHECK-NEXT: s_nop 1
; CHECK-NEXT: v_mov_b32_dpp v0, v0 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; CHECK-NEXT: s_setpc_b64 s[30:31]
%sum = fadd float %a, %b
%shuffle = call float @llvm.amdgcn.update.dpp.f32(
float poison, float %sum, i32 273, i32 15, i32 15, i1 true)
%result = call float @llvm.canonicalize.f32(float %shuffle)
ret float %result
}
define float @raw_f32_dpp(float %a) {
; CHECK-LABEL: raw_f32_dpp:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_mov_b32_dpp v0, v0 quad_perm:[1,0,3,2] row_mask:0xf bank_mask:0xf bound_ctrl:1
; CHECK-NEXT: v_max_f32_e32 v0, v0, v0
; CHECK-NEXT: s_setpc_b64 s[30:31]
%shuffle = call float @llvm.amdgcn.update.dpp.f32(
float poison, float %a, i32 177, i32 15, i32 15, i1 true)
%result = call float @llvm.canonicalize.f32(float %shuffle)
ret float %result
}
define float @canonical_dpp_partial_rows(float %a, float %b) {
; CHECK-LABEL: canonical_dpp_partial_rows:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_add_f32_e32 v0, v0, v1
; CHECK-NEXT: v_mov_b32_e32 v1, 0x7f800001
; CHECK-NEXT: s_nop 1
; CHECK-NEXT: v_mov_b32_dpp v1, v0 quad_perm:[1,0,3,2] row_mask:0x7 bank_mask:0xf bound_ctrl:1
; CHECK-NEXT: v_max_f32_e32 v0, v1, v1
; CHECK-NEXT: s_setpc_b64 s[30:31]
%sum = fadd float %a, %b
%bits = bitcast float %sum to i32
%shuffle = call i32 @llvm.amdgcn.update.dpp.i32(
i32 2139095041, i32 %bits, i32 177, i32 7, i32 15, i1 true)
%value = bitcast i32 %shuffle to float
%result = call float @llvm.canonicalize.f32(float %value)
ret float %result
}
define float @canonical_dpp_partial_banks(float %a, float %b) {
; CHECK-LABEL: canonical_dpp_partial_banks:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_add_f32_e32 v0, v0, v1
; CHECK-NEXT: v_mov_b32_e32 v1, 0x7f800001
; CHECK-NEXT: s_nop 1
; CHECK-NEXT: v_mov_b32_dpp v1, v0 quad_perm:[1,0,3,2] row_mask:0xf bank_mask:0x7 bound_ctrl:1
; CHECK-NEXT: v_max_f32_e32 v0, v1, v1
; CHECK-NEXT: s_setpc_b64 s[30:31]
%sum = fadd float %a, %b
%bits = bitcast float %sum to i32
%shuffle = call i32 @llvm.amdgcn.update.dpp.i32(
i32 2139095041, i32 %bits, i32 177, i32 15, i32 7, i1 true)
%value = bitcast i32 %shuffle to float
%result = call float @llvm.canonicalize.f32(float %value)
ret float %result
}
define float @canonical_dpp_without_zero_fill(float %a, float %b) {
; CHECK-LABEL: canonical_dpp_without_zero_fill:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_add_f32_e32 v0, v0, v1
; CHECK-NEXT: v_mov_b32_e32 v1, 0x7f800001
; CHECK-NEXT: s_nop 1
; CHECK-NEXT: v_mov_b32_dpp v1, v0 row_shr:1 row_mask:0xf bank_mask:0xf
; CHECK-NEXT: v_max_f32_e32 v0, v1, v1
; CHECK-NEXT: s_setpc_b64 s[30:31]
%sum = fadd float %a, %b
%bits = bitcast float %sum to i32
%shuffle = call i32 @llvm.amdgcn.update.dpp.i32(
i32 2139095041, i32 %bits, i32 273, i32 15, i32 15, i1 false)
%value = bitcast i32 %shuffle to float
%result = call float @llvm.canonicalize.f32(float %value)
ret float %result
}
define float @canonical_dpp_packed_f16_as_f32(
; CHECK-LABEL: canonical_dpp_packed_f16_as_f32:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_pk_add_f16 v0, v0, v1
; CHECK-NEXT: s_nop 1
; CHECK-NEXT: v_mov_b32_dpp v0, v0 quad_perm:[1,0,3,2] row_mask:0xf bank_mask:0xf bound_ctrl:1
; CHECK-NEXT: v_max_f32_e32 v0, v0, v0
; CHECK-NEXT: s_setpc_b64 s[30:31]
<2 x half> %a, <2 x half> %b) {
%sum = fadd <2 x half> %a, %b
%bits = bitcast <2 x half> %sum to i32
%shuffle = call i32 @llvm.amdgcn.update.dpp.i32(
i32 poison, i32 %bits, i32 177, i32 15, i32 15, i1 true)
%value = bitcast i32 %shuffle to float
%result = call float @llvm.canonicalize.f32(float %value)
ret float %result
}
declare i32 @llvm.amdgcn.update.dpp.i32(
i32, i32, i32 immarg, i32 immarg, i32 immarg, i1 immarg)
declare float @llvm.amdgcn.update.dpp.f32(
float, float, i32 immarg, i32 immarg, i32 immarg, i1 immarg)
declare float @llvm.canonicalize.f32(float)