intel/llvm
1
0
Fork 0
mirror of https://github.com/intel/llvm.git synced 2026-02-08 17:28:30 +08:00
Code Issues Packages Projects Releases Wiki Activity

AMDGPU/GlobalISel: Fully handle 0 dmask case during legalize

Browse Source 
For normal loads, fully eliminate the load. For the TFE case, adjust
the dmask value in the instruction so the selector doesn't need to
handle it. For the TFE special case, I guess it would be possible to
replace the loaded data register with undef, but as-is this will start
treating it as a well defined value.
This commit is contained in:
Matt Arsenault
2020-02-06 09:52:08 -05:00
committed by Matt Arsenault
parent d9a012ed8a
commit d0fe13ecf9
4 changed files with 83 additions and 130 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
View File

@@ -3549,7 +3549,22 @@ bool AMDGPULegalizerInfo::legalizeImageIntrinsic(
if (!BaseOpcode->Atomic) {
const int DMaskIdx = getDMaskIdx(BaseOpcode, NumDefs);
unsigned DMask = MI.getOperand(DMaskIdx).getImm();
DMaskLanes = BaseOpcode->Gather4 ? 4 : countPopulation(DMask);
if (BaseOpcode->Gather4) {
DMaskLanes = 4;
} else if (DMask != 0) {
DMaskLanes = countPopulation(DMask);
} else if (IsTFE) {
// Expecting to get an error flag since TFC is on - and dmask is 0 Force
// dmask to be at least 1 otherwise the instruction will fail
DMask = 0x1;
DMaskLanes = 1;
MI.getOperand(DMaskIdx).setImm(DMask);
} else if (!BaseOpcode->Store) {
// If dmask is 0, this is a no-op load. This can be eliminated.
B.buildUndef(MI.getOperand(0));
MI.eraseFromParent();
return true;
}
}
if (BaseOpcode->Store) { // No TFE for stores?

30
llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.dim.a16.ll
View File

@@ -2903,15 +2903,12 @@ define amdgpu_ps <4 x float> @getresinfo_dmask0(<8 x i32> inreg %rsrc, <4 x floa
; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; GFX9: [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 0, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; GFX9: $vgpr1 = COPY [[DEF]](s32)
; GFX9: $vgpr2 = COPY [[DEF]](s32)
; GFX9: $vgpr3 = COPY [[DEF]](s32)
; GFX9: [[DEF:%[0-9]+]]:_(<4 x s32>) = G_IMPLICIT_DEF
; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<4 x s32>)
; GFX9: $vgpr0 = COPY [[UV]](s32)
; GFX9: $vgpr1 = COPY [[UV1]](s32)
; GFX9: $vgpr2 = COPY [[UV2]](s32)
; GFX9: $vgpr3 = COPY [[UV3]](s32)
; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
; GFX10NSA-LABEL: name: getresinfo_dmask0
; GFX10NSA: bb.1.main_body:
@@ -2925,15 +2922,12 @@ define amdgpu_ps <4 x float> @getresinfo_dmask0(<8 x i32> inreg %rsrc, <4 x floa
; GFX10NSA: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
; GFX10NSA: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; GFX10NSA: [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 0, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; GFX10NSA: $vgpr1 = COPY [[DEF]](s32)
; GFX10NSA: $vgpr2 = COPY [[DEF]](s32)
; GFX10NSA: $vgpr3 = COPY [[DEF]](s32)
; GFX10NSA: [[DEF:%[0-9]+]]:_(<4 x s32>) = G_IMPLICIT_DEF
; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<4 x s32>)
; GFX10NSA: $vgpr0 = COPY [[UV]](s32)
; GFX10NSA: $vgpr1 = COPY [[UV1]](s32)
; GFX10NSA: $vgpr2 = COPY [[UV2]](s32)
; GFX10NSA: $vgpr3 = COPY [[UV3]](s32)
; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
main_body:
%mip = extractelement <2 x i16> %coords, i32 0

116
llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.d16.ll
View File

@@ -508,11 +508,8 @@ define amdgpu_ps half @image_load_f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s,
; UNPACKED: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; UNPACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s16) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
; UNPACKED: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[AMDGPU_INTRIN_IMAGE_LOAD]](s16)
; UNPACKED: $vgpr0 = COPY [[ANYEXT]](s32)
; UNPACKED: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; UNPACKED: $vgpr0 = COPY [[DEF]](s32)
; UNPACKED: SI_RETURN_TO_EPILOG implicit $vgpr0
; PACKED-LABEL: name: image_load_f16_dmask_0000
; PACKED: bb.1 (%ir-block.0):
@@ -527,11 +524,8 @@ define amdgpu_ps half @image_load_f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s,
; PACKED: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; PACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s16) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
; PACKED: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[AMDGPU_INTRIN_IMAGE_LOAD]](s16)
; PACKED: $vgpr0 = COPY [[ANYEXT]](s32)
; PACKED: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; PACKED: $vgpr0 = COPY [[DEF]](s32)
; PACKED: SI_RETURN_TO_EPILOG implicit $vgpr0
%tex = call half @llvm.amdgcn.image.load.2d.f16.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
ret half %tex
@@ -600,18 +594,8 @@ define amdgpu_ps <2 x half> @image_load_v2f16_dmask_0000(<8 x i32> inreg %rsrc,
; UNPACKED: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; UNPACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; UNPACKED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
; UNPACKED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
; UNPACKED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; UNPACKED: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
; UNPACKED: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
; UNPACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; UNPACKED: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
; UNPACKED: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
; UNPACKED: $vgpr0 = COPY [[DEF]](<2 x s16>)
; UNPACKED: SI_RETURN_TO_EPILOG implicit $vgpr0
; PACKED-LABEL: name: image_load_v2f16_dmask_0000
; PACKED: bb.1 (%ir-block.0):
@@ -626,10 +610,8 @@ define amdgpu_ps <2 x half> @image_load_v2f16_dmask_0000(<8 x i32> inreg %rsrc,
; PACKED: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; PACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; PACKED: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>)
; PACKED: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
; PACKED: $vgpr0 = COPY [[DEF]](<2 x s16>)
; PACKED: SI_RETURN_TO_EPILOG implicit $vgpr0
%tex = call <2 x half> @llvm.amdgcn.image.load.2d.v2f16.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
ret <2 x half> %tex
@@ -785,23 +767,10 @@ define amdgpu_ps <3 x half> @image_load_v3f16_dmask_0000(<8 x i32> inreg %rsrc,
; UNPACKED: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; UNPACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; UNPACKED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
; UNPACKED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
; UNPACKED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; UNPACKED: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
; UNPACKED: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
; UNPACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; UNPACKED: [[OR1:%[0-9]+]]:_(s32) = G_OR [[C1]], [[SHL]]
; UNPACKED: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; UNPACKED: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>)
; UNPACKED: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0
; UNPACKED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; UNPACKED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0
; UNPACKED: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0
; UNPACKED: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; UNPACKED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0
; UNPACKED: [[EXTRACT1:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
; UNPACKED: [[EXTRACT2:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 32
; UNPACKED: $vgpr0 = COPY [[EXTRACT1]](<2 x s16>)
@@ -820,18 +789,14 @@ define amdgpu_ps <3 x half> @image_load_v3f16_dmask_0000(<8 x i32> inreg %rsrc,
; PACKED: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; PACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
; PACKED: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
; PACKED: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>), [[DEF]](<2 x s16>), [[DEF]](<2 x s16>)
; PACKED: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
; PACKED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; PACKED: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0
; PACKED: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; PACKED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
; PACKED: [[EXTRACT:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
; PACKED: [[EXTRACT1:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 32
; PACKED: $vgpr0 = COPY [[EXTRACT]](<2 x s16>)
; PACKED: $vgpr1 = COPY [[EXTRACT1]](<2 x s16>)
; PACKED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0
; PACKED: [[EXTRACT1:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
; PACKED: [[EXTRACT2:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 32
; PACKED: $vgpr0 = COPY [[EXTRACT1]](<2 x s16>)
; PACKED: $vgpr1 = COPY [[EXTRACT2]](<2 x s16>)
; PACKED: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
%tex = call <3 x half> @llvm.amdgcn.image.load.2d.v3f16.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
ret <3 x half> %tex
@@ -1023,21 +988,10 @@ define amdgpu_ps <4 x half> @image_load_v4f16_dmask_0000(<8 x i32> inreg %rsrc,
; UNPACKED: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; UNPACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; UNPACKED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
; UNPACKED: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
; UNPACKED: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; UNPACKED: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
; UNPACKED: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
; UNPACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; UNPACKED: [[OR1:%[0-9]+]]:_(s32) = G_OR [[C1]], [[SHL]]
; UNPACKED: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; UNPACKED: $vgpr0 = COPY [[BITCAST]](<2 x s16>)
; UNPACKED: $vgpr1 = COPY [[BITCAST1]](<2 x s16>)
; UNPACKED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; UNPACKED: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF]](<4 x s16>)
; UNPACKED: $vgpr0 = COPY [[UV]](<2 x s16>)
; UNPACKED: $vgpr1 = COPY [[UV1]](<2 x s16>)
; UNPACKED: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
; PACKED-LABEL: name: image_load_v4f16_dmask_0000
; PACKED: bb.1 (%ir-block.0):
@@ -1052,12 +1006,10 @@ define amdgpu_ps <4 x half> @image_load_v4f16_dmask_0000(<8 x i32> inreg %rsrc,
; PACKED: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; PACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; PACKED: [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
; PACKED: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>)
; PACKED: $vgpr1 = COPY [[DEF]](<2 x s16>)
; PACKED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; PACKED: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF]](<4 x s16>)
; PACKED: $vgpr0 = COPY [[UV]](<2 x s16>)
; PACKED: $vgpr1 = COPY [[UV1]](<2 x s16>)
; PACKED: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
%tex = call <4 x half> @llvm.amdgcn.image.load.2d.v4f16.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
ret <4 x half> %tex
@@ -1080,7 +1032,7 @@ define amdgpu_ps half @image_load_tfe_f16_dmask_0000(<8 x i32> inreg %rsrc, i32
; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; UNPACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; UNPACKED: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
@@ -1102,7 +1054,7 @@ define amdgpu_ps half @image_load_tfe_f16_dmask_0000(<8 x i32> inreg %rsrc, i32
; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; PACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; PACKED: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
; PACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
@@ -1191,7 +1143,7 @@ define amdgpu_ps <2 x half> @image_load_tfe_v2f16_dmask_0000(<8 x i32> inreg %rs
; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; UNPACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
@@ -1220,7 +1172,7 @@ define amdgpu_ps <2 x half> @image_load_tfe_v2f16_dmask_0000(<8 x i32> inreg %rs
; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; PACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; PACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
; PACKED: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
@@ -1405,7 +1357,7 @@ define amdgpu_ps <3 x half> @image_load_tfe_v3f16_dmask_0000(<8 x i32> inreg %rs
; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; UNPACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
@@ -1443,7 +1395,7 @@ define amdgpu_ps <3 x half> @image_load_tfe_v3f16_dmask_0000(<8 x i32> inreg %rs
; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; PACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; PACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
; PACKED: [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
@@ -1681,7 +1633,7 @@ define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_0000(<8 x i32> inreg %rs
; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; UNPACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; UNPACKED: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
@@ -1713,7 +1665,7 @@ define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_0000(<8 x i32> inreg %rs
; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; PACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; PACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
; PACKED: [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF

50
llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.ll
View File

@@ -238,10 +238,8 @@ define amdgpu_ps float @image_load_f32_dmask_0000(<8 x i32> inreg %rsrc, i32 %s,
; GCN: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; GCN: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; GCN: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; GCN: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GCN: $vgpr0 = COPY [[DEF]](s32)
; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0
%tex = call float @llvm.amdgcn.image.load.2d.f32.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
ret float %tex
@@ -286,12 +284,10 @@ define amdgpu_ps <2 x float> @image_load_v2f32_dmask_0000(<8 x i32> inreg %rsrc,
; GCN: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; GCN: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; GCN: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GCN: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; GCN: $vgpr1 = COPY [[DEF]](s32)
; GCN: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
; GCN: $vgpr0 = COPY [[UV]](s32)
; GCN: $vgpr1 = COPY [[UV1]](s32)
; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
%tex = call <2 x float> @llvm.amdgcn.image.load.2d.v2f32.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
ret <2 x float> %tex
@@ -364,13 +360,11 @@ define amdgpu_ps <3 x float> @image_load_v3f32_dmask_0000(<8 x i32> inreg %rsrc,
; GCN: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; GCN: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
; GCN: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GCN: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; GCN: $vgpr1 = COPY [[DEF]](s32)
; GCN: $vgpr2 = COPY [[DEF]](s32)
; GCN: [[DEF:%[0-9]+]]:_(<3 x s32>) = G_IMPLICIT_DEF
; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<3 x s32>)
; GCN: $vgpr0 = COPY [[UV]](s32)
; GCN: $vgpr1 = COPY [[UV1]](s32)
; GCN: $vgpr2 = COPY [[UV2]](s32)
; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
%tex = call <3 x float> @llvm.amdgcn.image.load.2d.v3f32.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
ret <3 x float> %tex
@@ -473,14 +467,12 @@ define amdgpu_ps <4 x float> @image_load_v4f32_dmask_0000(<8 x i32> inreg %rsrc,
; GCN: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
; GCN: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
; GCN: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GCN: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
; GCN: $vgpr1 = COPY [[DEF]](s32)
; GCN: $vgpr2 = COPY [[DEF]](s32)
; GCN: $vgpr3 = COPY [[DEF]](s32)
; GCN: [[DEF:%[0-9]+]]:_(<4 x s32>) = G_IMPLICIT_DEF
; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<4 x s32>)
; GCN: $vgpr0 = COPY [[UV]](s32)
; GCN: $vgpr1 = COPY [[UV1]](s32)
; GCN: $vgpr2 = COPY [[UV2]](s32)
; GCN: $vgpr3 = COPY [[UV3]](s32)
; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
%tex = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
ret <4 x float> %tex
@@ -503,7 +495,7 @@ define amdgpu_ps float @image_load_tfe_f32_dmask_0000(<8 x i32> inreg %rsrc, i32
; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GCN: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; GCN: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
; GCN: $vgpr0 = COPY [[UV]](s32)
@@ -563,7 +555,7 @@ define amdgpu_ps <2 x float> @image_load_tfe_v2f32_dmask_0000(<8 x i32> inreg %r
; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GCN: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; GCN: [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GCN: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
@@ -658,7 +650,7 @@ define amdgpu_ps <3 x float> @image_load_tfe_v3f32_dmask_0000(<8 x i32> inreg %r
; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GCN: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; GCN: [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GCN: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
@@ -789,7 +781,7 @@ define amdgpu_ps <4 x float> @image_load_tfe_v4f32_dmask_0000(<8 x i32> inreg %r
; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
; GCN: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 0, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
; GCN: [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GCN: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)