blob: 0ef53be350707eb700b8677ef51817d934a28bef [file] [edit]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -mtriple=amdgpu9.42 < %s | FileCheck -check-prefixes=GFX9X,GFX942,GFX942-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgpu9.42 < %s | FileCheck -check-prefixes=GFX9X,GFX942,GFX942-GISEL %s
; RUN: llc -global-isel=0 -mtriple=amdgpu9.50 < %s | FileCheck -check-prefixes=GFX9X,GFX950,GFX950-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgpu9.50 < %s | FileCheck -check-prefixes=GFX9X,GFX950,GFX950-GISEL %s
; RUN: llc -global-isel=0 -mtriple=amdgpu11.70 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX1170,GFX1170-TRUE16 %s
; RUN: llc -global-isel=0 -mtriple=amdgpu11.70 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1170,GFX1170-FAKE16 %s
; RUN: llc -global-isel=1 -mtriple=amdgpu11.70 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1170,GFX1170-GISEL %s
; RUN: llc -global-isel=0 -mtriple=amdgpu12.00 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-TRUE16 %s
; RUN: llc -global-isel=0 -mtriple=amdgpu12.00 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-FAKE16 %s
; RUN: llc -global-isel=1 -mtriple=amdgpu12.00 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
; RUN: llc -global-isel=0 -mtriple=amdgpu12.50 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-TRUE16 %s
; RUN: llc -global-isel=0 -mtriple=amdgpu12.50 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-FAKE16 %s
; RUN: llc -global-isel=1 -mtriple=amdgpu12.50 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s
declare float @llvm.amdgcn.cvt.f32.bf8(i32, i32)
declare float @llvm.amdgcn.cvt.f32.fp8(i32, i32)
declare <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32, i1)
declare <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32, i1)
declare i32 @llvm.amdgcn.cvt.pk.bf8.f32(float, float, i32, i1)
declare i32 @llvm.amdgcn.cvt.pk.fp8.f32(float, float, i32, i1)
declare i32 @llvm.amdgcn.cvt.sr.bf8.f32(float, i32, i32, i32)
declare i32 @llvm.amdgcn.cvt.sr.fp8.f32(float, i32, i32, i32)
define float @test_cvt_f32_bf8_byte0(i32 %a) {
; GFX942-LABEL: test_cvt_f32_bf8_byte0:
; GFX942: ; %bb.0:
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX942-NEXT: v_cvt_f32_bf8_sdwa v0, v0 src0_sel:BYTE_0
; GFX942-NEXT: s_setpc_b64 s[30:31]
;
; GFX950-LABEL: test_cvt_f32_bf8_byte0:
; GFX950: ; %bb.0:
; GFX950-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX950-NEXT: v_cvt_f32_bf8_e32 v0, v0
; GFX950-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_f32_bf8_byte0:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_f32_bf8_e32 v0, v0
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_f32_bf8_byte0:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_f32_bf8_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_f32_bf8_byte0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_f32_bf8_e32 v0, v0
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %a, i32 0)
ret float %ret
}
define float @test_cvt_f32_bf8_byte1(i32 %a) {
; GFX9X-LABEL: test_cvt_f32_bf8_byte1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_f32_bf8_sdwa v0, v0 src0_sel:BYTE_1
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_f32_bf8_byte1:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:1
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_f32_bf8_byte1:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:1
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_f32_bf8_byte1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:1
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %a, i32 1)
ret float %ret
}
define float @test_cvt_f32_bf8_byte2(i32 %a) {
; GFX9X-LABEL: test_cvt_f32_bf8_byte2:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_f32_bf8_sdwa v0, v0 src0_sel:BYTE_2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_f32_bf8_byte2:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_f32_bf8_byte2:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_f32_bf8_byte2:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %a, i32 2)
ret float %ret
}
define float @test_cvt_f32_bf8_byte3(i32 %a) {
; GFX9X-LABEL: test_cvt_f32_bf8_byte3:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_f32_bf8_sdwa v0, v0 src0_sel:BYTE_3
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_f32_bf8_byte3:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:3
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_f32_bf8_byte3:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:3
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_f32_bf8_byte3:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:3
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %a, i32 3)
ret float %ret
}
define float @test_cvt_f32_fp8_byte0(i32 %a) {
; GFX942-LABEL: test_cvt_f32_fp8_byte0:
; GFX942: ; %bb.0:
; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX942-NEXT: v_cvt_f32_fp8_sdwa v0, v0 src0_sel:BYTE_0
; GFX942-NEXT: s_setpc_b64 s[30:31]
;
; GFX950-LABEL: test_cvt_f32_fp8_byte0:
; GFX950: ; %bb.0:
; GFX950-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX950-NEXT: v_cvt_f32_fp8_e32 v0, v0
; GFX950-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_f32_fp8_byte0:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_f32_fp8_e32 v0, v0
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_f32_fp8_byte0:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_f32_fp8_e32 v0, v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_f32_fp8_byte0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_f32_fp8_e32 v0, v0
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call float @llvm.amdgcn.cvt.f32.fp8(i32 %a, i32 0)
ret float %ret
}
define float @test_cvt_f32_fp8_byte1(i32 %a) {
; GFX9X-LABEL: test_cvt_f32_fp8_byte1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_f32_fp8_sdwa v0, v0 src0_sel:BYTE_1
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_f32_fp8_byte1:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:1
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_f32_fp8_byte1:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:1
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_f32_fp8_byte1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:1
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call float @llvm.amdgcn.cvt.f32.fp8(i32 %a, i32 1)
ret float %ret
}
define float @test_cvt_f32_fp8_byte2(i32 %a) {
; GFX9X-LABEL: test_cvt_f32_fp8_byte2:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_f32_fp8_sdwa v0, v0 src0_sel:BYTE_2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_f32_fp8_byte2:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_f32_fp8_byte2:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_f32_fp8_byte2:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call float @llvm.amdgcn.cvt.f32.fp8(i32 %a, i32 2)
ret float %ret
}
define float @test_cvt_f32_fp8_byte3(i32 %a) {
; GFX9X-LABEL: test_cvt_f32_fp8_byte3:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_f32_fp8_sdwa v0, v0 src0_sel:BYTE_3
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_f32_fp8_byte3:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:3
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_f32_fp8_byte3:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:3
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_f32_fp8_byte3:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:3
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call float @llvm.amdgcn.cvt.f32.fp8(i32 %a, i32 3)
ret float %ret
}
define <2 x float> @test_cvt_pk_f32_bf8_word0(i32 %a) {
; GFX9X-LABEL: test_cvt_pk_f32_bf8_word0:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_pk_f32_bf8_e32 v[0:1], v0
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_pk_f32_bf8_word0:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_pk_f32_bf8_e32 v[0:1], v0
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_pk_f32_bf8_word0:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_pk_f32_bf8_e32 v[0:1], v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_pk_f32_bf8_word0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_pk_f32_bf8_e32 v[0:1], v0
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32 %a, i1 false)
ret <2 x float> %ret
}
define <2 x float> @test_cvt_pk_f32_bf8_word1(i32 %a) {
; GFX9X-LABEL: test_cvt_pk_f32_bf8_word1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_pk_f32_bf8_sdwa v[0:1], v0 src0_sel:WORD_1
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_pk_f32_bf8_word1:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_pk_f32_bf8_word1:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_pk_f32_bf8_word1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32 %a, i1 true)
ret <2 x float> %ret
}
define <2 x float> @test_cvt_pk_f32_fp8_word0(i32 %a) {
; GFX9X-LABEL: test_cvt_pk_f32_fp8_word0:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], v0
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_pk_f32_fp8_word0:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], v0
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_pk_f32_fp8_word0:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_pk_f32_fp8_word0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], v0
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32 %a, i1 false)
ret <2 x float> %ret
}
define <2 x float> @test_cvt_pk_f32_fp8_word1(i32 %a) {
; GFX9X-LABEL: test_cvt_pk_f32_fp8_word1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_pk_f32_fp8_sdwa v[0:1], v0 src0_sel:WORD_1
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_pk_f32_fp8_word1:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_pk_f32_fp8_word1:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_pk_f32_fp8_word1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32 %a, i1 true)
ret <2 x float> %ret
}
define i32 @test_cvt_pk_bf8_f32_word0(float %x, float %y, i32 %old) {
; GFX9X-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-TRUE16-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX1170-TRUE16: ; %bb.0:
; GFX1170-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-TRUE16-NEXT: v_cvt_pk_bf8_f32 v2.l, v0, v1
; GFX1170-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-FAKE16-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX1170-FAKE16: ; %bb.0:
; GFX1170-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-FAKE16-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1
; GFX1170-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-GISEL-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX1170-GISEL: ; %bb.0:
; GFX1170-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-GISEL-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1
; GFX1170-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX12-TRUE16: ; %bb.0:
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_cvt_pk_bf8_f32 v2.l, v0, v1
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX12-FAKE16: ; %bb.0:
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT: v_cvt_pk_bf8_f32 v2.l, v0, v1
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-TRUE16-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX1250-TRUE16: ; %bb.0:
; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-TRUE16-NEXT: v_cvt_pk_bf8_f32 v2.l, v0, v1
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31]
;
; GFX1250-FAKE16-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31]
;
; GFX1250-GISEL-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX1250-GISEL: ; %bb.0:
; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX1250-GISEL-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-GISEL-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.pk.bf8.f32(float %x, float %y, i32 %old, i1 false)
ret i32 %ret
}
define i32 @test_cvt_pk_bf8_f32_word1(float %x, float %y, i32 %old) {
; GFX9X-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX9X-NEXT: s_nop 0
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-TRUE16-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX1170-TRUE16: ; %bb.0:
; GFX1170-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-TRUE16-NEXT: v_cvt_pk_bf8_f32 v2.h, v0, v1
; GFX1170-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-FAKE16-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX1170-FAKE16: ; %bb.0:
; GFX1170-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-FAKE16-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1170-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-GISEL-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX1170-GISEL: ; %bb.0:
; GFX1170-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-GISEL-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1170-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX12-TRUE16: ; %bb.0:
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_cvt_pk_bf8_f32 v2.h, v0, v1
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX12-FAKE16: ; %bb.0:
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT: v_cvt_pk_bf8_f32 v2.h, v0, v1
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-TRUE16-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX1250-TRUE16: ; %bb.0:
; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-TRUE16-NEXT: v_cvt_pk_bf8_f32 v2.h, v0, v1
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31]
;
; GFX1250-FAKE16-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31]
;
; GFX1250-GISEL-LABEL: test_cvt_pk_bf8_f32_word1:
; GFX1250-GISEL: ; %bb.0:
; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX1250-GISEL-NEXT: v_cvt_pk_bf8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-GISEL-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.pk.bf8.f32(float %x, float %y, i32 %old, i1 true)
ret i32 %ret
}
define i32 @test_cvt_pk_fp8_f32_word0(float %x, float %y, i32 %old) {
; GFX9X-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-TRUE16-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX1170-TRUE16: ; %bb.0:
; GFX1170-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-TRUE16-NEXT: v_cvt_pk_fp8_f32 v2.l, v0, v1
; GFX1170-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-FAKE16-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX1170-FAKE16: ; %bb.0:
; GFX1170-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-FAKE16-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1
; GFX1170-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-GISEL-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX1170-GISEL: ; %bb.0:
; GFX1170-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-GISEL-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1
; GFX1170-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX12-TRUE16: ; %bb.0:
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_cvt_pk_fp8_f32 v2.l, v0, v1
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX12-FAKE16: ; %bb.0:
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT: v_cvt_pk_fp8_f32 v2.l, v0, v1
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-TRUE16-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX1250-TRUE16: ; %bb.0:
; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-TRUE16-NEXT: v_cvt_pk_fp8_f32 v2.l, v0, v1
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31]
;
; GFX1250-FAKE16-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31]
;
; GFX1250-GISEL-LABEL: test_cvt_pk_fp8_f32_word0:
; GFX1250-GISEL: ; %bb.0:
; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX1250-GISEL-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-GISEL-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.pk.fp8.f32(float %x, float %y, i32 %old, i1 false)
ret i32 %ret
}
define i32 @test_cvt_pk_fp8_f32_word1(float %x, float %y, i32 %old) {
; GFX9X-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX9X-NEXT: s_nop 0
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-TRUE16-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX1170-TRUE16: ; %bb.0:
; GFX1170-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-TRUE16-NEXT: v_cvt_pk_fp8_f32 v2.h, v0, v1
; GFX1170-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-FAKE16-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX1170-FAKE16: ; %bb.0:
; GFX1170-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-FAKE16-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1170-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-GISEL-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX1170-GISEL: ; %bb.0:
; GFX1170-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-GISEL-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1170-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-TRUE16-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX12-TRUE16: ; %bb.0:
; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX12-TRUE16-NEXT: v_cvt_pk_fp8_f32 v2.h, v0, v1
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-FAKE16-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX12-FAKE16: ; %bb.0:
; GFX12-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-FAKE16-NEXT: s_wait_expcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX12-FAKE16-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT: v_cvt_pk_fp8_f32 v2.h, v0, v1
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-TRUE16-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX1250-TRUE16: ; %bb.0:
; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-TRUE16-NEXT: v_cvt_pk_fp8_f32 v2.h, v0, v1
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31]
;
; GFX1250-FAKE16-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31]
;
; GFX1250-GISEL-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX1250-GISEL: ; %bb.0:
; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX1250-GISEL-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-GISEL-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.pk.fp8.f32(float %x, float %y, i32 %old, i1 true)
ret i32 %ret
}
define i32 @test_cvt_sr_bf8_f32_byte0(float %x, i32 %r, i32 %old) {
; GFX9X-LABEL: test_cvt_sr_bf8_f32_byte0:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_sr_bf8_f32_byte0:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_sr_bf8_f32_byte0:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, v2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_sr_bf8_f32_byte0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.sr.bf8.f32(float %x, i32 %r, i32 %old, i32 0)
ret i32 %ret
}
define i32 @test_cvt_sr_bf8_f32_byte1(float %x, i32 %r, i32 %old) {
; GFX9X-LABEL: test_cvt_sr_bf8_f32_byte1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 op_sel:[0,0,1,0]
; GFX9X-NEXT: s_nop 0
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_sr_bf8_f32_byte1:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:1
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_sr_bf8_f32_byte1:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:1
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, v2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_sr_bf8_f32_byte1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:1
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.sr.bf8.f32(float %x, i32 %r, i32 %old, i32 1)
ret i32 %ret
}
define i32 @test_cvt_sr_bf8_f32_byte2(float %x, i32 %r, i32 %old) {
; GFX9X-LABEL: test_cvt_sr_bf8_f32_byte2:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 op_sel:[0,0,0,1]
; GFX9X-NEXT: s_nop 0
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_sr_bf8_f32_byte2:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:2
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_sr_bf8_f32_byte2:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:2
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, v2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_sr_bf8_f32_byte2:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:2
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.sr.bf8.f32(float %x, i32 %r, i32 %old, i32 2)
ret i32 %ret
}
define i32 @test_cvt_sr_bf8_f32_byte3(float %x, i32 %r, i32 %old) {
; GFX9X-LABEL: test_cvt_sr_bf8_f32_byte3:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 op_sel:[0,0,1,1]
; GFX9X-NEXT: s_nop 0
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_sr_bf8_f32_byte3:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:3
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_sr_bf8_f32_byte3:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:3
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, v2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_sr_bf8_f32_byte3:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_sr_bf8_f32 v2, v0, v1 byte_sel:3
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.sr.bf8.f32(float %x, i32 %r, i32 %old, i32 3)
ret i32 %ret
}
define i32 @test_cvt_sr_fp8_f32_byte0(float %x, i32 %r, i32 %old) {
; GFX9X-LABEL: test_cvt_sr_fp8_f32_byte0:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_sr_fp8_f32_byte0:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_sr_fp8_f32_byte0:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, v2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_sr_fp8_f32_byte0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.sr.fp8.f32(float %x, i32 %r, i32 %old, i32 0)
ret i32 %ret
}
define i32 @test_cvt_sr_fp8_f32_byte1(float %x, i32 %r, i32 %old) {
; GFX9X-LABEL: test_cvt_sr_fp8_f32_byte1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 op_sel:[0,0,1,0]
; GFX9X-NEXT: s_nop 0
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_sr_fp8_f32_byte1:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:1
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_sr_fp8_f32_byte1:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:1
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, v2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_sr_fp8_f32_byte1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:1
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.sr.fp8.f32(float %x, i32 %r, i32 %old, i32 1)
ret i32 %ret
}
define i32 @test_cvt_sr_fp8_f32_byte2(float %x, i32 %r, i32 %old) {
; GFX9X-LABEL: test_cvt_sr_fp8_f32_byte2:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 op_sel:[0,0,0,1]
; GFX9X-NEXT: s_nop 0
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_sr_fp8_f32_byte2:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:2
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_sr_fp8_f32_byte2:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:2
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, v2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_sr_fp8_f32_byte2:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:2
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.sr.fp8.f32(float %x, i32 %r, i32 %old, i32 2)
ret i32 %ret
}
define i32 @test_cvt_sr_fp8_f32_byte3(float %x, i32 %r, i32 %old) {
; GFX9X-LABEL: test_cvt_sr_fp8_f32_byte3:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 op_sel:[0,0,1,1]
; GFX9X-NEXT: s_nop 0
; GFX9X-NEXT: v_mov_b32_e32 v0, v2
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_cvt_sr_fp8_f32_byte3:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:3
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_mov_b32_e32 v0, v2
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_cvt_sr_fp8_f32_byte3:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:3
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, v2
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_cvt_sr_fp8_f32_byte3:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_cvt_sr_fp8_f32 v2, v0, v1 byte_sel:3
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_mov_b32_e32 v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%ret = tail call i32 @llvm.amdgcn.cvt.sr.fp8.f32(float %x, i32 %r, i32 %old, i32 3)
ret i32 %ret
}
define float @test_sext_cvt_f32_fp8(i16 %a) {
; GFX9X-LABEL: test_sext_cvt_f32_fp8:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX9X-NEXT: v_cvt_f32_fp8_sdwa v0, v0 src0_sel:BYTE_1
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_sext_cvt_f32_fp8:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:1
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_sext_cvt_f32_fp8:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:1
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_sext_cvt_f32_fp8:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_cvt_f32_fp8_e64 v0, v0 byte_sel:1
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%a.sext = sext i16 %a to i32
%ret = tail call float @llvm.amdgcn.cvt.f32.fp8(i32 %a.sext, i32 1)
ret float %ret
}
define float @test_sext_cvt_f32_bf8(i16 %a) {
; GFX9X-LABEL: test_sext_cvt_f32_bf8:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX9X-NEXT: v_cvt_f32_bf8_sdwa v0, v0 src0_sel:BYTE_1
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_sext_cvt_f32_bf8:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:1
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_sext_cvt_f32_bf8:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:1
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_sext_cvt_f32_bf8:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_cvt_f32_bf8_e64 v0, v0 byte_sel:1
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%a.sext = sext i16 %a to i32
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %a.sext, i32 1)
ret float %ret
}
define <2 x float> @test_sext_cvt_pk_f32_bf8_word1(i16 %a) {
; GFX9X-LABEL: test_sext_cvt_pk_f32_bf8_word1:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX9X-NEXT: v_cvt_pk_f32_bf8_sdwa v[0:1], v0 src0_sel:WORD_1
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_sext_cvt_pk_f32_bf8_word1:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_sext_cvt_pk_f32_bf8_word1:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_sext_cvt_pk_f32_bf8_word1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%a.sext = sext i16 %a to i32
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32 %a.sext, i1 true)
ret <2 x float> %ret
}
define <2 x float> @test_sext_cvt_pk_f32_fp8_word0(i16 %a) {
; GFX9X-LABEL: test_sext_cvt_pk_f32_fp8_word0:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9X-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX9X-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], v0
; GFX9X-NEXT: s_setpc_b64 s[30:31]
;
; GFX1170-LABEL: test_sext_cvt_pk_f32_fp8_word0:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1170-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1170-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], v0
; GFX1170-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: test_sext_cvt_pk_f32_fp8_word0:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], v0
; GFX12-NEXT: s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: test_sext_cvt_pk_f32_fp8_word0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_bfe_i32 v0, v0, 0, 16
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], v0
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%a.sext = sext i16 %a to i32
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32 %a.sext, i1 false)
ret <2 x float> %ret
}
define amdgpu_ps <2 x float> @test_cvt_pk_f32_bf8_word0_s(i32 inreg %a) {
; GFX9X-LABEL: test_cvt_pk_f32_bf8_word0_s:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: v_cvt_pk_f32_bf8_e32 v[0:1], s0
; GFX9X-NEXT: ; return to shader part epilog
;
; GFX1170-LABEL: test_cvt_pk_f32_bf8_word0_s:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: v_cvt_pk_f32_bf8_e32 v[0:1], s0
; GFX1170-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: test_cvt_pk_f32_bf8_word0_s:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_pk_f32_bf8_e32 v[0:1], s0
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_pk_f32_bf8_word0_s:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-NEXT: v_nop
; GFX1250-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-NEXT: v_cvt_pk_f32_bf8_e32 v[0:1], s0
; GFX1250-NEXT: ; return to shader part epilog
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32 %a, i1 false)
ret <2 x float> %ret
}
define amdgpu_ps <2 x float> @test_cvt_pk_f32_bf8_word1_s(i32 inreg %a) {
; GFX942-SDAG-LABEL: test_cvt_pk_f32_bf8_word1_s:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_cvt_pk_f32_bf8_sdwa v[0:1], s0 src0_sel:WORD_1
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_pk_f32_bf8_word1_s:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX942-GISEL-NEXT: v_cvt_pk_f32_bf8_sdwa v[0:1], v0 src0_sel:WORD_1
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-SDAG-LABEL: test_cvt_pk_f32_bf8_word1_s:
; GFX950-SDAG: ; %bb.0:
; GFX950-SDAG-NEXT: v_cvt_pk_f32_bf8_sdwa v[0:1], s0 src0_sel:WORD_1
; GFX950-SDAG-NEXT: ; return to shader part epilog
;
; GFX950-GISEL-LABEL: test_cvt_pk_f32_bf8_word1_s:
; GFX950-GISEL: ; %bb.0:
; GFX950-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX950-GISEL-NEXT: v_cvt_pk_f32_bf8_sdwa v[0:1], v0 src0_sel:WORD_1
; GFX950-GISEL-NEXT: ; return to shader part epilog
;
; GFX1170-LABEL: test_cvt_pk_f32_bf8_word1_s:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], s0 op_sel:[1,0]
; GFX1170-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: test_cvt_pk_f32_bf8_word1_s:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], s0 op_sel:[1,0]
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_pk_f32_bf8_word1_s:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-NEXT: v_nop
; GFX1250-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-NEXT: v_cvt_pk_f32_bf8_e64 v[0:1], s0 op_sel:[1,0]
; GFX1250-NEXT: ; return to shader part epilog
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32 %a, i1 true)
ret <2 x float> %ret
}
define amdgpu_ps <2 x float> @test_cvt_pk_f32_fp8_word0_s(i32 inreg %a) {
; GFX9X-LABEL: test_cvt_pk_f32_fp8_word0_s:
; GFX9X: ; %bb.0:
; GFX9X-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], s0
; GFX9X-NEXT: ; return to shader part epilog
;
; GFX1170-LABEL: test_cvt_pk_f32_fp8_word0_s:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], s0
; GFX1170-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: test_cvt_pk_f32_fp8_word0_s:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], s0
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_pk_f32_fp8_word0_s:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-NEXT: v_nop
; GFX1250-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-NEXT: v_cvt_pk_f32_fp8_e32 v[0:1], s0
; GFX1250-NEXT: ; return to shader part epilog
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32 %a, i1 false)
ret <2 x float> %ret
}
define amdgpu_ps <2 x float> @test_cvt_pk_f32_fp8_word1_s(i32 inreg %a) {
; GFX942-SDAG-LABEL: test_cvt_pk_f32_fp8_word1_s:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_cvt_pk_f32_fp8_sdwa v[0:1], s0 src0_sel:WORD_1
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_pk_f32_fp8_word1_s:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX942-GISEL-NEXT: v_cvt_pk_f32_fp8_sdwa v[0:1], v0 src0_sel:WORD_1
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-SDAG-LABEL: test_cvt_pk_f32_fp8_word1_s:
; GFX950-SDAG: ; %bb.0:
; GFX950-SDAG-NEXT: v_cvt_pk_f32_fp8_sdwa v[0:1], s0 src0_sel:WORD_1
; GFX950-SDAG-NEXT: ; return to shader part epilog
;
; GFX950-GISEL-LABEL: test_cvt_pk_f32_fp8_word1_s:
; GFX950-GISEL: ; %bb.0:
; GFX950-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX950-GISEL-NEXT: v_cvt_pk_f32_fp8_sdwa v[0:1], v0 src0_sel:WORD_1
; GFX950-GISEL-NEXT: ; return to shader part epilog
;
; GFX1170-LABEL: test_cvt_pk_f32_fp8_word1_s:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: v_cvt_pk_f32_fp8_e64 v[0:1], s0 op_sel:[1,0]
; GFX1170-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: test_cvt_pk_f32_fp8_word1_s:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_pk_f32_fp8_e64 v[0:1], s0 op_sel:[1,0]
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_pk_f32_fp8_word1_s:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-NEXT: v_nop
; GFX1250-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-NEXT: v_cvt_pk_f32_fp8_e64 v[0:1], s0 op_sel:[1,0]
; GFX1250-NEXT: ; return to shader part epilog
%ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32 %a, i1 true)
ret <2 x float> %ret
}
define amdgpu_ps float @test_cvt_f32_bf8_byte0_s(i32 inreg %a) {
; GFX942-SDAG-LABEL: test_cvt_f32_bf8_byte0_s:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_cvt_f32_bf8_sdwa v0, s0 src0_sel:BYTE_0
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_f32_bf8_byte0_s:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX942-GISEL-NEXT: v_cvt_f32_bf8_sdwa v0, v0 src0_sel:BYTE_0
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-LABEL: test_cvt_f32_bf8_byte0_s:
; GFX950: ; %bb.0:
; GFX950-NEXT: v_cvt_f32_bf8_e32 v0, s0
; GFX950-NEXT: ; return to shader part epilog
;
; GFX1170-LABEL: test_cvt_f32_bf8_byte0_s:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: v_cvt_f32_bf8_e32 v0, s0
; GFX1170-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: test_cvt_f32_bf8_byte0_s:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_f32_bf8_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_f32_bf8_byte0_s:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-NEXT: v_nop
; GFX1250-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-NEXT: v_cvt_f32_bf8_e32 v0, s0
; GFX1250-NEXT: ; return to shader part epilog
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %a, i32 0)
ret float %ret
}
define amdgpu_ps float @test_cvt_f32_fp8_byte0_s(i32 inreg %a) {
; GFX942-SDAG-LABEL: test_cvt_f32_fp8_byte0_s:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_cvt_f32_fp8_sdwa v0, s0 src0_sel:BYTE_0
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_f32_fp8_byte0_s:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX942-GISEL-NEXT: v_cvt_f32_fp8_sdwa v0, v0 src0_sel:BYTE_0
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-LABEL: test_cvt_f32_fp8_byte0_s:
; GFX950: ; %bb.0:
; GFX950-NEXT: v_cvt_f32_fp8_e32 v0, s0
; GFX950-NEXT: ; return to shader part epilog
;
; GFX1170-LABEL: test_cvt_f32_fp8_byte0_s:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: v_cvt_f32_fp8_e32 v0, s0
; GFX1170-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: test_cvt_f32_fp8_byte0_s:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_f32_fp8_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_f32_fp8_byte0_s:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-NEXT: v_nop
; GFX1250-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-NEXT: v_cvt_f32_fp8_e32 v0, s0
; GFX1250-NEXT: ; return to shader part epilog
%ret = tail call float @llvm.amdgcn.cvt.f32.fp8(i32 %a, i32 0)
ret float %ret
}
define amdgpu_ps i32 @test_cvt_pk_bf8_f32_word0_sss(float inreg %x, float inreg %y, i32 inreg %old) {
; GFX942-SDAG-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX942-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX942-SDAG-NEXT: v_cvt_pk_bf8_f32 v0, s0, v1
; GFX942-SDAG-NEXT: s_nop 0
; GFX942-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX942-GISEL-NEXT: v_cvt_pk_bf8_f32 v1, s0, v0
; GFX942-GISEL-NEXT: s_nop 0
; GFX942-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-SDAG-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX950-SDAG: ; %bb.0:
; GFX950-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX950-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX950-SDAG-NEXT: v_cvt_pk_bf8_f32 v0, s0, v1
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX950-SDAG-NEXT: ; return to shader part epilog
;
; GFX950-GISEL-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX950-GISEL: ; %bb.0:
; GFX950-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX950-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX950-GISEL-NEXT: v_cvt_pk_bf8_f32 v1, s0, v0
; GFX950-GISEL-NEXT: s_nop 0
; GFX950-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX950-GISEL-NEXT: ; return to shader part epilog
;
; GFX1170-TRUE16-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX1170-TRUE16: ; %bb.0:
; GFX1170-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-TRUE16-NEXT: v_cvt_pk_bf8_f32 v0.l, s0, s1
; GFX1170-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-TRUE16-NEXT: ; return to shader part epilog
;
; GFX1170-FAKE16-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX1170-FAKE16: ; %bb.0:
; GFX1170-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-FAKE16-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1
; GFX1170-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-FAKE16-NEXT: ; return to shader part epilog
;
; GFX1170-GISEL-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX1170-GISEL: ; %bb.0:
; GFX1170-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-GISEL-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1
; GFX1170-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-GISEL-NEXT: ; return to shader part epilog
;
; GFX12-TRUE16-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX12-TRUE16: ; %bb.0:
; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-TRUE16-NEXT: v_cvt_pk_bf8_f32 v0.l, s0, s1
; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-TRUE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-TRUE16-NEXT: ; return to shader part epilog
;
; GFX12-FAKE16-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX12-FAKE16: ; %bb.0:
; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1
; GFX12-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-FAKE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-FAKE16-NEXT: ; return to shader part epilog
;
; GFX12-GISEL-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_cvt_pk_bf8_f32 v0.l, s0, s1
; GFX12-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-GISEL-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-GISEL-NEXT: ; return to shader part epilog
;
; GFX1250-TRUE16-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX1250-TRUE16: ; %bb.0:
; GFX1250-TRUE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-TRUE16-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-TRUE16-NEXT: v_nop
; GFX1250-TRUE16-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_cvt_pk_bf8_f32 v0.l, s0, s1
; GFX1250-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-TRUE16-NEXT: ; return to shader part epilog
;
; GFX1250-FAKE16-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-FAKE16-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-FAKE16-NEXT: v_nop
; GFX1250-FAKE16-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1
; GFX1250-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-FAKE16-NEXT: ; return to shader part epilog
;
; GFX1250-GISEL-LABEL: test_cvt_pk_bf8_f32_word0_sss:
; GFX1250-GISEL: ; %bb.0:
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-GISEL-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-GISEL-NEXT: ; return to shader part epilog
%ret = tail call i32 @llvm.amdgcn.cvt.pk.bf8.f32(float %x, float %y, i32 %old, i1 false)
ret i32 %ret
}
define amdgpu_ps i32 @test_cvt_pk_bf8_f32_word1_sss(float inreg %x, float inreg %y, i32 inreg %old) {
; GFX942-SDAG-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX942-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX942-SDAG-NEXT: v_cvt_pk_bf8_f32 v0, s0, v1 op_sel:[0,0,1]
; GFX942-SDAG-NEXT: s_nop 0
; GFX942-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX942-GISEL-NEXT: v_cvt_pk_bf8_f32 v1, s0, v0 op_sel:[0,0,1]
; GFX942-GISEL-NEXT: s_nop 0
; GFX942-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-SDAG-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX950-SDAG: ; %bb.0:
; GFX950-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX950-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX950-SDAG-NEXT: v_cvt_pk_bf8_f32 v0, s0, v1 op_sel:[0,0,1]
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX950-SDAG-NEXT: ; return to shader part epilog
;
; GFX950-GISEL-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX950-GISEL: ; %bb.0:
; GFX950-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX950-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX950-GISEL-NEXT: v_cvt_pk_bf8_f32 v1, s0, v0 op_sel:[0,0,1]
; GFX950-GISEL-NEXT: s_nop 0
; GFX950-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX950-GISEL-NEXT: ; return to shader part epilog
;
; GFX1170-TRUE16-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX1170-TRUE16: ; %bb.0:
; GFX1170-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-TRUE16-NEXT: v_cvt_pk_bf8_f32 v0.h, s0, s1
; GFX1170-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-TRUE16-NEXT: ; return to shader part epilog
;
; GFX1170-FAKE16-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX1170-FAKE16: ; %bb.0:
; GFX1170-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-FAKE16-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX1170-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-FAKE16-NEXT: ; return to shader part epilog
;
; GFX1170-GISEL-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX1170-GISEL: ; %bb.0:
; GFX1170-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-GISEL-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX1170-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-GISEL-NEXT: ; return to shader part epilog
;
; GFX12-TRUE16-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX12-TRUE16: ; %bb.0:
; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-TRUE16-NEXT: v_cvt_pk_bf8_f32 v0.h, s0, s1
; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-TRUE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-TRUE16-NEXT: ; return to shader part epilog
;
; GFX12-FAKE16-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX12-FAKE16: ; %bb.0:
; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX12-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-FAKE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-FAKE16-NEXT: ; return to shader part epilog
;
; GFX12-GISEL-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_cvt_pk_bf8_f32 v0.h, s0, s1
; GFX12-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-GISEL-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-GISEL-NEXT: ; return to shader part epilog
;
; GFX1250-TRUE16-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX1250-TRUE16: ; %bb.0:
; GFX1250-TRUE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-TRUE16-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-TRUE16-NEXT: v_nop
; GFX1250-TRUE16-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_cvt_pk_bf8_f32 v0.h, s0, s1
; GFX1250-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-TRUE16-NEXT: ; return to shader part epilog
;
; GFX1250-FAKE16-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-FAKE16-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-FAKE16-NEXT: v_nop
; GFX1250-FAKE16-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX1250-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-FAKE16-NEXT: ; return to shader part epilog
;
; GFX1250-GISEL-LABEL: test_cvt_pk_bf8_f32_word1_sss:
; GFX1250-GISEL: ; %bb.0:
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-GISEL-NEXT: v_cvt_pk_bf8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-GISEL-NEXT: ; return to shader part epilog
%ret = tail call i32 @llvm.amdgcn.cvt.pk.bf8.f32(float %x, float %y, i32 %old, i1 true)
ret i32 %ret
}
define amdgpu_ps i32 @test_cvt_pk_fp8_f32_word0_sss(float inreg %x, float inreg %y, i32 inreg %old) {
; GFX942-SDAG-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX942-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX942-SDAG-NEXT: v_cvt_pk_fp8_f32 v0, s0, v1
; GFX942-SDAG-NEXT: s_nop 0
; GFX942-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX942-GISEL-NEXT: v_cvt_pk_fp8_f32 v1, s0, v0
; GFX942-GISEL-NEXT: s_nop 0
; GFX942-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-SDAG-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX950-SDAG: ; %bb.0:
; GFX950-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX950-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX950-SDAG-NEXT: v_cvt_pk_fp8_f32 v0, s0, v1
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX950-SDAG-NEXT: ; return to shader part epilog
;
; GFX950-GISEL-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX950-GISEL: ; %bb.0:
; GFX950-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX950-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX950-GISEL-NEXT: v_cvt_pk_fp8_f32 v1, s0, v0
; GFX950-GISEL-NEXT: s_nop 0
; GFX950-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX950-GISEL-NEXT: ; return to shader part epilog
;
; GFX1170-TRUE16-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX1170-TRUE16: ; %bb.0:
; GFX1170-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-TRUE16-NEXT: v_cvt_pk_fp8_f32 v0.l, s0, s1
; GFX1170-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-TRUE16-NEXT: ; return to shader part epilog
;
; GFX1170-FAKE16-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX1170-FAKE16: ; %bb.0:
; GFX1170-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-FAKE16-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1
; GFX1170-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-FAKE16-NEXT: ; return to shader part epilog
;
; GFX1170-GISEL-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX1170-GISEL: ; %bb.0:
; GFX1170-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-GISEL-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1
; GFX1170-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-GISEL-NEXT: ; return to shader part epilog
;
; GFX12-TRUE16-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX12-TRUE16: ; %bb.0:
; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-TRUE16-NEXT: v_cvt_pk_fp8_f32 v0.l, s0, s1
; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-TRUE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-TRUE16-NEXT: ; return to shader part epilog
;
; GFX12-FAKE16-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX12-FAKE16: ; %bb.0:
; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1
; GFX12-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-FAKE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-FAKE16-NEXT: ; return to shader part epilog
;
; GFX12-GISEL-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_cvt_pk_fp8_f32 v0.l, s0, s1
; GFX12-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-GISEL-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-GISEL-NEXT: ; return to shader part epilog
;
; GFX1250-TRUE16-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX1250-TRUE16: ; %bb.0:
; GFX1250-TRUE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-TRUE16-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-TRUE16-NEXT: v_nop
; GFX1250-TRUE16-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_cvt_pk_fp8_f32 v0.l, s0, s1
; GFX1250-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-TRUE16-NEXT: ; return to shader part epilog
;
; GFX1250-FAKE16-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-FAKE16-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-FAKE16-NEXT: v_nop
; GFX1250-FAKE16-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1
; GFX1250-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-FAKE16-NEXT: ; return to shader part epilog
;
; GFX1250-GISEL-LABEL: test_cvt_pk_fp8_f32_word0_sss:
; GFX1250-GISEL: ; %bb.0:
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-GISEL-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-GISEL-NEXT: ; return to shader part epilog
%ret = tail call i32 @llvm.amdgcn.cvt.pk.fp8.f32(float %x, float %y, i32 %old, i1 false)
ret i32 %ret
}
define amdgpu_ps i32 @test_cvt_pk_fp8_f32_word1_sss(float inreg %x, float inreg %y, i32 inreg %old) {
; GFX942-SDAG-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX942-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX942-SDAG-NEXT: v_cvt_pk_fp8_f32 v0, s0, v1 op_sel:[0,0,1]
; GFX942-SDAG-NEXT: s_nop 0
; GFX942-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX942-GISEL-NEXT: v_cvt_pk_fp8_f32 v1, s0, v0 op_sel:[0,0,1]
; GFX942-GISEL-NEXT: s_nop 0
; GFX942-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-SDAG-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX950-SDAG: ; %bb.0:
; GFX950-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX950-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX950-SDAG-NEXT: v_cvt_pk_fp8_f32 v0, s0, v1 op_sel:[0,0,1]
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX950-SDAG-NEXT: ; return to shader part epilog
;
; GFX950-GISEL-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX950-GISEL: ; %bb.0:
; GFX950-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX950-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX950-GISEL-NEXT: v_cvt_pk_fp8_f32 v1, s0, v0 op_sel:[0,0,1]
; GFX950-GISEL-NEXT: s_nop 0
; GFX950-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX950-GISEL-NEXT: ; return to shader part epilog
;
; GFX1170-TRUE16-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX1170-TRUE16: ; %bb.0:
; GFX1170-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-TRUE16-NEXT: v_cvt_pk_fp8_f32 v0.h, s0, s1
; GFX1170-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-TRUE16-NEXT: ; return to shader part epilog
;
; GFX1170-FAKE16-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX1170-FAKE16: ; %bb.0:
; GFX1170-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-FAKE16-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX1170-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-FAKE16-NEXT: ; return to shader part epilog
;
; GFX1170-GISEL-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX1170-GISEL: ; %bb.0:
; GFX1170-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-GISEL-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX1170-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-GISEL-NEXT: ; return to shader part epilog
;
; GFX12-TRUE16-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX12-TRUE16: ; %bb.0:
; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-TRUE16-NEXT: v_cvt_pk_fp8_f32 v0.h, s0, s1
; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-TRUE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-TRUE16-NEXT: ; return to shader part epilog
;
; GFX12-FAKE16-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX12-FAKE16: ; %bb.0:
; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-FAKE16-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX12-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-FAKE16-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-FAKE16-NEXT: ; return to shader part epilog
;
; GFX12-GISEL-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_cvt_pk_fp8_f32 v0.h, s0, s1
; GFX12-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-GISEL-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-GISEL-NEXT: ; return to shader part epilog
;
; GFX1250-TRUE16-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX1250-TRUE16: ; %bb.0:
; GFX1250-TRUE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-TRUE16-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-TRUE16-NEXT: v_nop
; GFX1250-TRUE16-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_cvt_pk_fp8_f32 v0.h, s0, s1
; GFX1250-TRUE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-TRUE16-NEXT: ; return to shader part epilog
;
; GFX1250-FAKE16-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-FAKE16-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-FAKE16-NEXT: v_nop
; GFX1250-FAKE16-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX1250-FAKE16-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-FAKE16-NEXT: ; return to shader part epilog
;
; GFX1250-GISEL-LABEL: test_cvt_pk_fp8_f32_word1_sss:
; GFX1250-GISEL: ; %bb.0:
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-GISEL-NEXT: v_cvt_pk_fp8_f32 v0, s0, s1 op_sel:[0,0,1]
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-GISEL-NEXT: ; return to shader part epilog
%ret = tail call i32 @llvm.amdgcn.cvt.pk.fp8.f32(float %x, float %y, i32 %old, i1 true)
ret i32 %ret
}
define amdgpu_ps i32 @test_cvt_sr_bf8_f32_byte0_sss(float inreg %x, i32 inreg %r, i32 inreg %old) {
; GFX942-SDAG-LABEL: test_cvt_sr_bf8_f32_byte0_sss:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX942-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX942-SDAG-NEXT: v_cvt_sr_bf8_f32 v0, s0, v1
; GFX942-SDAG-NEXT: s_nop 0
; GFX942-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_sr_bf8_f32_byte0_sss:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX942-GISEL-NEXT: v_cvt_sr_bf8_f32 v1, s0, v0
; GFX942-GISEL-NEXT: s_nop 0
; GFX942-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-SDAG-LABEL: test_cvt_sr_bf8_f32_byte0_sss:
; GFX950-SDAG: ; %bb.0:
; GFX950-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX950-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX950-SDAG-NEXT: v_cvt_sr_bf8_f32 v0, s0, v1
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX950-SDAG-NEXT: ; return to shader part epilog
;
; GFX950-GISEL-LABEL: test_cvt_sr_bf8_f32_byte0_sss:
; GFX950-GISEL: ; %bb.0:
; GFX950-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX950-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX950-GISEL-NEXT: v_cvt_sr_bf8_f32 v1, s0, v0
; GFX950-GISEL-NEXT: s_nop 0
; GFX950-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX950-GISEL-NEXT: ; return to shader part epilog
;
; GFX1170-LABEL: test_cvt_sr_bf8_f32_byte0_sss:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-NEXT: v_cvt_sr_bf8_f32 v0, s0, s1
; GFX1170-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: test_cvt_sr_bf8_f32_byte0_sss:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_mov_b32_e32 v0, s2
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_cvt_sr_bf8_f32 v0, s0, s1
; GFX12-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_sr_bf8_f32_byte0_sss:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-NEXT: v_nop
; GFX1250-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-NEXT: v_cvt_sr_bf8_f32 v0, s0, s1
; GFX1250-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-NEXT: ; return to shader part epilog
%ret = tail call i32 @llvm.amdgcn.cvt.sr.bf8.f32(float %x, i32 %r, i32 %old, i32 0)
ret i32 %ret
}
define amdgpu_ps i32 @test_cvt_sr_fp8_f32_byte0_sss(float inreg %x, i32 inreg %r, i32 inreg %old) {
; GFX942-SDAG-LABEL: test_cvt_sr_fp8_f32_byte0_sss:
; GFX942-SDAG: ; %bb.0:
; GFX942-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX942-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX942-SDAG-NEXT: v_cvt_sr_fp8_f32 v0, s0, v1
; GFX942-SDAG-NEXT: s_nop 0
; GFX942-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX942-SDAG-NEXT: ; return to shader part epilog
;
; GFX942-GISEL-LABEL: test_cvt_sr_fp8_f32_byte0_sss:
; GFX942-GISEL: ; %bb.0:
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX942-GISEL-NEXT: v_cvt_sr_fp8_f32 v1, s0, v0
; GFX942-GISEL-NEXT: s_nop 0
; GFX942-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX942-GISEL-NEXT: ; return to shader part epilog
;
; GFX950-SDAG-LABEL: test_cvt_sr_fp8_f32_byte0_sss:
; GFX950-SDAG: ; %bb.0:
; GFX950-SDAG-NEXT: v_mov_b32_e32 v0, s2
; GFX950-SDAG-NEXT: v_mov_b32_e32 v1, s1
; GFX950-SDAG-NEXT: v_cvt_sr_fp8_f32 v0, s0, v1
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX950-SDAG-NEXT: ; return to shader part epilog
;
; GFX950-GISEL-LABEL: test_cvt_sr_fp8_f32_byte0_sss:
; GFX950-GISEL: ; %bb.0:
; GFX950-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX950-GISEL-NEXT: v_mov_b32_e32 v1, s2
; GFX950-GISEL-NEXT: v_cvt_sr_fp8_f32 v1, s0, v0
; GFX950-GISEL-NEXT: s_nop 0
; GFX950-GISEL-NEXT: v_readfirstlane_b32 s0, v1
; GFX950-GISEL-NEXT: ; return to shader part epilog
;
; GFX1170-LABEL: test_cvt_sr_fp8_f32_byte0_sss:
; GFX1170: ; %bb.0:
; GFX1170-NEXT: v_mov_b32_e32 v0, s2
; GFX1170-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1170-NEXT: v_cvt_sr_fp8_f32 v0, s0, s1
; GFX1170-NEXT: v_readfirstlane_b32 s0, v0
; GFX1170-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: test_cvt_sr_fp8_f32_byte0_sss:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_mov_b32_e32 v0, s2
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX12-NEXT: v_cvt_sr_fp8_f32 v0, s0, s1
; GFX12-NEXT: v_readfirstlane_b32 s0, v0
; GFX12-NEXT: s_wait_alu depctr_va_sdst(0)
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_sr_fp8_f32_byte0_sss:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-NEXT: v_nop
; GFX1250-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-NEXT: v_mov_b32_e32 v0, s2
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-NEXT: v_cvt_sr_fp8_f32 v0, s0, s1
; GFX1250-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-NEXT: ; return to shader part epilog
%ret = tail call i32 @llvm.amdgcn.cvt.sr.fp8.f32(float %x, i32 %r, i32 %old, i32 0)
ret i32 %ret
}