f2c164c815
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
271 lines
12 KiB
LLVM
271 lines
12 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
|
|
; FIXME: Missing operand promote for f16
|
|
; 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 float @v_constrained_fpext_f16_to_f32_fpexcept_strict(half %arg) #0 {
|
|
; GCN-LABEL: v_constrained_fpext_f16_to_f32_fpexcept_strict:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call float @llvm.experimental.constrained.fpext.f32.f16(half %arg, metadata !"fpexcept.strict")
|
|
ret float %result
|
|
}
|
|
|
|
define <2 x float> @v_constrained_fpext_v2f16_to_v2f32_fpexcept_strict(<2 x half> %arg) #0 {
|
|
; GFX89-LABEL: v_constrained_fpext_v2f16_to_v2f32_fpexcept_strict:
|
|
; GFX89: ; %bb.0:
|
|
; GFX89-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX89-NEXT: v_cvt_f32_f16_e32 v2, v0
|
|
; GFX89-NEXT: v_cvt_f32_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
|
|
; GFX89-NEXT: v_mov_b32_e32 v0, v2
|
|
; GFX89-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX10-LABEL: v_constrained_fpext_v2f16_to_v2f32_fpexcept_strict:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v2, v0
|
|
; GFX10-NEXT: v_cvt_f32_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, v2
|
|
; GFX10-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_constrained_fpext_v2f16_to_v2f32_fpexcept_strict:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v1, v1
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call <2 x float> @llvm.experimental.constrained.fpext.v2f32.v2f16(<2 x half> %arg, metadata !"fpexcept.strict")
|
|
ret <2 x float> %result
|
|
}
|
|
|
|
define <3 x float> @v_constrained_fpext_v3f16_to_v3f32_fpexcept_strict(<3 x half> %arg) #0 {
|
|
; GFX89-LABEL: v_constrained_fpext_v3f16_to_v3f32_fpexcept_strict:
|
|
; GFX89: ; %bb.0:
|
|
; GFX89-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX89-NEXT: v_cvt_f32_f16_e32 v4, v0
|
|
; GFX89-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
|
|
; GFX89-NEXT: v_cvt_f32_f16_e32 v2, v1
|
|
; GFX89-NEXT: v_mov_b32_e32 v0, v4
|
|
; GFX89-NEXT: v_mov_b32_e32 v1, v3
|
|
; GFX89-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX10-LABEL: v_constrained_fpext_v3f16_to_v3f32_fpexcept_strict:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v4, v0
|
|
; GFX10-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v2, v1
|
|
; GFX10-NEXT: v_mov_b32_e32 v0, v4
|
|
; GFX10-NEXT: v_mov_b32_e32 v1, v3
|
|
; GFX10-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_constrained_fpext_v3f16_to_v3f32_fpexcept_strict:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_lshrrev_b32_e32 v2, 16, v0
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v2
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v1
|
|
; GFX11-NEXT: v_mov_b32_e32 v1, v3
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call <3 x float> @llvm.experimental.constrained.fpext.v3f32.v3f16(<3 x half> %arg, metadata !"fpexcept.strict")
|
|
ret <3 x float> %result
|
|
}
|
|
|
|
define double @v_constrained_fpext_f32_to_f64_fpexcept_strict(float %arg) #0 {
|
|
; GCN-LABEL: v_constrained_fpext_f32_to_f64_fpexcept_strict:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call double @llvm.experimental.constrained.fpext.f64.f32(float %arg, metadata !"fpexcept.strict")
|
|
ret double %result
|
|
}
|
|
|
|
define <2 x double> @v_constrained_fpext_v2f32_to_v2f64_fpexcept_strict(<2 x float> %arg) #0 {
|
|
; GCN-LABEL: v_constrained_fpext_v2f32_to_v2f64_fpexcept_strict:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_mov_b32_e32 v2, v1
|
|
; GCN-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GCN-NEXT: v_cvt_f64_f32_e32 v[2:3], v2
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call <2 x double> @llvm.experimental.constrained.fpext.v2f64.v2f32(<2 x float> %arg, metadata !"fpexcept.strict")
|
|
ret <2 x double> %result
|
|
}
|
|
|
|
define <3 x double> @v_constrained_fpext_v3f32_to_v3f64_fpexcept_strict(<3 x float> %arg) #0 {
|
|
; GFX89-LABEL: v_constrained_fpext_v3f32_to_v3f64_fpexcept_strict:
|
|
; GFX89: ; %bb.0:
|
|
; GFX89-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX89-NEXT: v_mov_b32_e32 v4, v2
|
|
; GFX89-NEXT: v_mov_b32_e32 v2, v1
|
|
; GFX89-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GFX89-NEXT: v_cvt_f64_f32_e32 v[2:3], v2
|
|
; GFX89-NEXT: v_cvt_f64_f32_e32 v[4:5], v4
|
|
; GFX89-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX1011-LABEL: v_constrained_fpext_v3f32_to_v3f64_fpexcept_strict:
|
|
; GFX1011: ; %bb.0:
|
|
; GFX1011-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX1011-NEXT: v_mov_b32_e32 v4, v2
|
|
; GFX1011-NEXT: v_mov_b32_e32 v2, v1
|
|
; GFX1011-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GFX1011-NEXT: v_cvt_f64_f32_e32 v[4:5], v4
|
|
; GFX1011-NEXT: v_cvt_f64_f32_e32 v[2:3], v2
|
|
; GFX1011-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call <3 x double> @llvm.experimental.constrained.fpext.v3f64.v3f32(<3 x float> %arg, metadata !"fpexcept.strict")
|
|
ret <3 x double> %result
|
|
}
|
|
|
|
define double @v_constrained_fpext_f16_to_f64_fpexcept_strict(half %arg) #0 {
|
|
; GCN-LABEL: v_constrained_fpext_f16_to_f64_fpexcept_strict:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; GCN-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call double @llvm.experimental.constrained.fpext.f64.f16(half %arg, metadata !"fpexcept.strict")
|
|
ret double %result
|
|
}
|
|
|
|
define <2 x double> @v_constrained_fpext_v2f16_to_v2f64_fpexcept_strict(<2 x half> %arg) #0 {
|
|
; GFX89-LABEL: v_constrained_fpext_v2f16_to_v2f64_fpexcept_strict:
|
|
; GFX89: ; %bb.0:
|
|
; GFX89-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX89-NEXT: v_cvt_f32_f16_e32 v1, v0
|
|
; GFX89-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
|
|
; GFX89-NEXT: v_cvt_f64_f32_e32 v[0:1], v1
|
|
; GFX89-NEXT: v_cvt_f64_f32_e32 v[2:3], v2
|
|
; GFX89-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX10-LABEL: v_constrained_fpext_v2f16_to_v2f64_fpexcept_strict:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v1, v0
|
|
; GFX10-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
|
|
; GFX10-NEXT: v_cvt_f64_f32_e32 v[0:1], v1
|
|
; GFX10-NEXT: v_cvt_f64_f32_e32 v[2:3], v2
|
|
; GFX10-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_constrained_fpext_v2f16_to_v2f64_fpexcept_strict:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v1
|
|
; GFX11-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GFX11-NEXT: v_cvt_f64_f32_e32 v[2:3], v2
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call <2 x double> @llvm.experimental.constrained.fpext.v2f64.v2f16(<2 x half> %arg, metadata !"fpexcept.strict")
|
|
ret <2 x double> %result
|
|
}
|
|
|
|
define <3 x double> @v_constrained_fpext_v3f16_to_v2f64_fpexcept_strict(<3 x half> %arg) #0 {
|
|
; GFX89-LABEL: v_constrained_fpext_v3f16_to_v2f64_fpexcept_strict:
|
|
; GFX89: ; %bb.0:
|
|
; GFX89-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX89-NEXT: v_cvt_f32_f16_e32 v2, v0
|
|
; GFX89-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
|
|
; GFX89-NEXT: v_cvt_f32_f16_e32 v4, v1
|
|
; GFX89-NEXT: v_cvt_f64_f32_e32 v[0:1], v2
|
|
; GFX89-NEXT: v_cvt_f64_f32_e32 v[2:3], v3
|
|
; GFX89-NEXT: v_cvt_f64_f32_e32 v[4:5], v4
|
|
; GFX89-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX10-LABEL: v_constrained_fpext_v3f16_to_v2f64_fpexcept_strict:
|
|
; GFX10: ; %bb.0:
|
|
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v2, v0
|
|
; GFX10-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
|
|
; GFX10-NEXT: v_cvt_f32_f16_e32 v4, v1
|
|
; GFX10-NEXT: v_cvt_f64_f32_e32 v[0:1], v2
|
|
; GFX10-NEXT: v_cvt_f64_f32_e32 v[2:3], v3
|
|
; GFX10-NEXT: v_cvt_f64_f32_e32 v[4:5], v4
|
|
; GFX10-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: v_constrained_fpext_v3f16_to_v2f64_fpexcept_strict:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: v_lshrrev_b32_e32 v2, 16, v0
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
|
|
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v2
|
|
; GFX11-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GFX11-NEXT: v_cvt_f64_f32_e32 v[4:5], v3
|
|
; GFX11-NEXT: v_cvt_f64_f32_e32 v[2:3], v2
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call <3 x double> @llvm.experimental.constrained.fpext.v3f64.v3f16(<3 x half> %arg, metadata !"fpexcept.strict")
|
|
ret <3 x double> %result
|
|
}
|
|
|
|
define float @v_constrained_fneg_fpext_f16_to_f32_fpexcept_strict(half %arg) #0 {
|
|
; GCN-LABEL: v_constrained_fneg_fpext_f16_to_f32_fpexcept_strict:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_cvt_f32_f16_e32 v0, v0
|
|
; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call float @llvm.experimental.constrained.fpext.f32.f16(half %arg, metadata !"fpexcept.strict")
|
|
%neg.result = fneg float %result
|
|
ret float %neg.result
|
|
}
|
|
|
|
define float @v_constrained_fpext_fneg_f16_to_f32_fpexcept_strict(half %arg) #0 {
|
|
; GCN-LABEL: v_constrained_fpext_fneg_f16_to_f32_fpexcept_strict:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_cvt_f32_f16_e64 v0, -v0
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
%neg.arg = fneg half %arg
|
|
%result = call float @llvm.experimental.constrained.fpext.f32.f16(half %neg.arg, metadata !"fpexcept.strict")
|
|
ret float %result
|
|
}
|
|
|
|
define double @v_constrained_fpext_fneg_f32_to_f64_fpexcept_strict(float %arg) #0 {
|
|
; GCN-LABEL: v_constrained_fpext_fneg_f32_to_f64_fpexcept_strict:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
%neg.arg = fneg float %arg
|
|
%result = call double @llvm.experimental.constrained.fpext.f64.f32(float %arg, metadata !"fpexcept.strict")
|
|
ret double %result
|
|
}
|
|
|
|
define double @v_constrained_fneg_fpext_f32_to_f64_fpexcept_strict(float %arg) #0 {
|
|
; GCN-LABEL: v_constrained_fneg_fpext_f32_to_f64_fpexcept_strict:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_cvt_f64_f32_e32 v[0:1], v0
|
|
; GCN-NEXT: v_xor_b32_e32 v1, 0x80000000, v1
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
%result = call double @llvm.experimental.constrained.fpext.f64.f32(float %arg, metadata !"fpexcept.strict")
|
|
%neg.result = fneg double %result
|
|
ret double %neg.result
|
|
}
|
|
|
|
declare double @llvm.experimental.constrained.fpext.f64.f32(float, metadata) #1
|
|
declare <2 x double> @llvm.experimental.constrained.fpext.v2f64.v2f32(<2 x float>, metadata) #1
|
|
declare <3 x double> @llvm.experimental.constrained.fpext.v3f64.v3f32(<3 x float>, metadata) #1
|
|
|
|
declare double @llvm.experimental.constrained.fpext.f64.f16(half, metadata) #1
|
|
declare <2 x double> @llvm.experimental.constrained.fpext.v2f64.v2f16(<2 x half>, metadata) #1
|
|
declare <3 x double> @llvm.experimental.constrained.fpext.v3f64.v3f16(<3 x half>, metadata) #1
|
|
|
|
declare float @llvm.experimental.constrained.fpext.f32.f16(half, metadata) #1
|
|
declare <2 x float> @llvm.experimental.constrained.fpext.v2f32.v2f16(<2 x half>, metadata) #1
|
|
declare <3 x float> @llvm.experimental.constrained.fpext.v3f32.v3f16(<3 x half>, 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:
|
|
; GFX8: {{.*}}
|
|
; GFX9: {{.*}}
|