caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
1daf2994de49d1ecba4bee4e6842aa8a564cbc96
clang-p2996/llvm/test/CodeGen/AMDGPU/strict_fptrunc.ll
Jay Foad f2c164c815 [AMDGPU] Do not wait for vscnt on function entry and return 
SIInsertWaitcnts inserts waitcnt instructions to resolve data
dependencies. The GFX10+ vscnt (VMEM store count) counter is never used
in this way. It is only used to resolve memory dependencies, and that is
handled by SIMemoryLegalizer. Hence there is no need to conservatively
wait for vscnt to be 0 on function entry and before returns.

Differential Revision: https://reviews.llvm.org/D153537
2023-07-04 12:22:38 +01:00

192 lines
9.3 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; XUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,SI %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx803 < %s | FileCheck -check-prefixes=GCN,GFX89,GFX8 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX89,GFX9 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GCN,GFX1011,GFX10 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GCN,GFX1011,GFX11 %s
define half @v_constrained_fptrunc_f32_to_f16_fpexcept_strict(float %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_cvt_f16_f32_e32 v0, v0
; GCN-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
define <2 x half> @v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict(<2 x float> %arg) #0 {
; GFX8-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: v_cvt_f16_f32_sdwa v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_or_b32_e32 v0, v0, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX1011-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; GFX1011: ; %bb.0:
; GFX1011-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1011-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX1011-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX1011-NEXT: v_perm_b32 v0, v1, v0, 0x5040100
; GFX1011-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f32(<2 x float> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
define <3 x half> @v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict(<3 x float> %arg) #0 {
; GFX8-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_cvt_f16_f32_sdwa v3, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: v_cvt_f16_f32_e32 v1, v2
; GFX8-NEXT: v_or_b32_e32 v0, v0, v3
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT: v_cvt_f16_f32_e32 v3, v1
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v2
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
; GFX9-NEXT: v_perm_b32 v0, v3, v0, s4
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX1011-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; GFX1011: ; %bb.0:
; GFX1011-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1011-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX1011-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX1011-NEXT: v_perm_b32 v0, v1, v0, 0x5040100
; GFX1011-NEXT: v_cvt_f16_f32_e32 v1, v2
; GFX1011-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x half> @llvm.experimental.constrained.fptrunc.v3f16.v3f32(<3 x float> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x half> %val
}
define float @v_constrained_fptrunc_f64_to_f32_fpexcept_strict(double %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_f64_to_f32_fpexcept_strict:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_cvt_f32_f64_e32 v0, v[0:1]
; GCN-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fptrunc.f32.f64(double %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
define <2 x float> @v_constrained_fptrunc_v2f64_to_v2f32_fpexcept_strict(<2 x double> %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_v2f64_to_v2f32_fpexcept_strict:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_cvt_f32_f64_e32 v0, v[0:1]
; GCN-NEXT: v_cvt_f32_f64_e32 v1, v[2:3]
; GCN-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fptrunc.v2f32.v2f64(<2 x double> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x float> %val
}
define <3 x float> @v_constrained_fptrunc_v3f64_to_v3f32_fpexcept_strict(<3 x double> %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_v3f64_to_v3f32_fpexcept_strict:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_cvt_f32_f64_e32 v0, v[0:1]
; GCN-NEXT: v_cvt_f32_f64_e32 v1, v[2:3]
; GCN-NEXT: v_cvt_f32_f64_e32 v2, v[4:5]
; GCN-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x float> @llvm.experimental.constrained.fptrunc.v3f32.v3f64(<3 x double> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x float> %val
}
; FIXME:
; define half @v_constrained_fptrunc_f64_to_f16_fpexcept_strict(double %arg) #0 {
; %val = call half @llvm.experimental.constrained.fptrunc.f16.f64(double %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
; ret half %val
; }
; define <2 x half> @v_constrained_fptrunc_v2f64_to_v2f16_fpexcept_strict(<2 x double> %arg) #0 {
; %val = call <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f64(<2 x double> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
; ret <2 x half> %val
; }
; define <3 x half> @v_constrained_fptrunc_v3f64_to_v3f16_fpexcept_strict(<3 x double> %arg) #0 {
; %val = call <3 x half> @llvm.experimental.constrained.fptrunc.v3f16.v3f64(<3 x double> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
; ret <3 x half> %val
; }
define half @v_constrained_fneg_fptrunc_f32_to_f16_fpexcept_strict(float %arg) #0 {
; GCN-LABEL: v_constrained_fneg_fptrunc_f32_to_f16_fpexcept_strict:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_cvt_f16_f32_e32 v0, v0
; GCN-NEXT: v_xor_b32_e32 v0, 0x8000, v0
; GCN-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
%neg.val = fneg half %val
ret half %neg.val
}
define half @v_constrained_fptrunc_fneg_f32_to_f16_fpexcept_strict(float %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_fneg_f32_to_f16_fpexcept_strict:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_cvt_f16_f32_e64 v0, -v0
; GCN-NEXT: s_setpc_b64 s[30:31]
%neg.arg = fneg float %arg
%val = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %neg.arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
define float @v_constrained_fneg_fptrunc_f64_to_f32_fpexcept_strict(double %arg) #0 {
; GCN-LABEL: v_constrained_fneg_fptrunc_f64_to_f32_fpexcept_strict:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_cvt_f32_f64_e32 v0, v[0:1]
; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0
; GCN-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fptrunc.f32.f64(double %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
%neg.val = fneg float %val
ret float %neg.val
}
define float @v_constrained_fptrunc_fneg_f64_to_f32_fpexcept_strict(double %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_fneg_f64_to_f32_fpexcept_strict:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_cvt_f32_f64_e64 v0, -v[0:1]
; GCN-NEXT: s_setpc_b64 s[30:31]
%neg.arg = fneg double %arg
%val = call float @llvm.experimental.constrained.fptrunc.f32.f64(double %neg.arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
declare half @llvm.experimental.constrained.fptrunc.f16.f32(float, metadata, metadata) #1
declare <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f32(<2 x float>, metadata, metadata) #1
declare <3 x half> @llvm.experimental.constrained.fptrunc.v3f16.v3f32(<3 x float>, metadata, metadata) #1
declare float @llvm.experimental.constrained.fptrunc.f32.f64(double, metadata, metadata) #1
declare <2 x float> @llvm.experimental.constrained.fptrunc.v2f32.v2f64(<2 x double>, metadata, metadata) #1
declare <3 x float> @llvm.experimental.constrained.fptrunc.v3f32.v3f64(<3 x double>, metadata, metadata) #1
declare half @llvm.experimental.constrained.fptrunc.f16.f64(double, metadata, metadata) #1
declare <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f64(<2 x double>, metadata, metadata) #1
declare <3 x half> @llvm.experimental.constrained.fptrunc.v3f16.v3f64(<3 x double>, metadata, metadata) #1
attributes #0 = { strictfp }
attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GFX10: {{.*}}
; GFX11: {{.*}}
; GFX89: {{.*}}
Reference in New Issue View Git Blame Copy Permalink