[llvm-branch-commits] [llvm] AMDGPU: Promote 16-bit arguments to i32 in the calling convention (PR #242694)
via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Sat Nov 14 11:26:11 PST 2026
llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Matt Arsenault (arsenm)
<details>
<summary>Changes</summary>
Promote i16, f16 and bf16 arguments to i32 in the calling convention
when they are inreg or real true16 is not used, instead of special
casing the copy type in LowerFormalArguments. The manual call lowering
code can now rely on LocVT / ArgVT.
Co-authored-by: Claude Opus 5.5 <noreply@<!-- -->anthropic.com>
---
Patch is 404.31 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/242694.diff
15 Files Affected:
- (modified) llvm/lib/Target/AMDGPU/AMDGPUCallingConv.td (+9)
- (modified) llvm/lib/Target/AMDGPU/SIISelLowering.cpp (+17-13)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/and.ll (+8-38)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-implicit-args.ll (+34-68)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call.ll (+6-3)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-function-args.ll (+53-83)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/or.ll (+8-38)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/xor.ll (+8-38)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.1024bit.ll (+231-231)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.256bit.ll (+604-1128)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.320bit.ll (+752-1412)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.512bit.ll (+342-342)
- (modified) llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll (+4-4)
- (modified) llvm/test/CodeGen/AMDGPU/isel-amdgpu-cs-chain-cc.ll (+12-18)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.set.rounding.ll (+106-314)
``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCallingConv.td b/llvm/lib/Target/AMDGPU/AMDGPUCallingConv.td
index 157c8541a78c56..926fa12f529afa 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUCallingConv.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPUCallingConv.td
@@ -16,6 +16,9 @@ class CCIfExtend<CCAction A>
: CCIf<"ArgFlags.isSExt() || ArgFlags.isZExt()", A>;
class CCIfOrigTypeShaderCCIsSGPR<CCAction A>
: CCIf<[{!OrigTy->getScalarType()->isFloatingPointTy()}], A>;
+class CCIfInRegOrNoRealTrue16<CCAction A>
+ : CCIf<"ArgFlags.isInReg() || !State.getMachineFunction()"
+ ".getSubtarget<GCNSubtarget>().useRealTrue16Insts()", A>;
// Calling convention for SI
@@ -25,6 +28,8 @@ def CC_SI_Gfx : CallingConv<[
// 32 is reserved for the stack pointer
// 33 is reserved for the frame pointer
// 34 is reserved for the base pointer
+ CCIfType<[i16, f16, bf16], CCIfInRegOrNoRealTrue16<CCPromoteToType<i32>>>,
+
CCIfInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16, bf16, v2bf16] , CCAssignToReg<
!foreach(i, !range(4, 30), !cast<Register>("SGPR"#i)) // SGPR4-29
>>>,
@@ -48,6 +53,7 @@ def RetCC_SI_Gfx : CallingConv<[
def CC_SI_SHADER : CallingConv<[
CCIfType<[i1], CCPromoteToType<i32>>,
+ CCIfType<[i16, f16, bf16], CCIfInRegOrNoRealTrue16<CCPromoteToType<i32>>>,
CCIfInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16, bf16, v2bf16] , CCAssignToReg<
!foreach(i, !range(0, 44), !cast<Register>("SGPR"#i)) // SGPR0-43
@@ -137,6 +143,7 @@ def CC_AMDGPU_Func : CallingConv<[
CCIfByVal<CCPassByVal<4, 4>>,
CCIfType<[i1], CCPromoteToType<i32>>,
CCIfType<[i8, i16], CCIfExtend<CCPromoteToType<i32>>>,
+ CCIfType<[i16, f16, bf16], CCIfInRegOrNoRealTrue16<CCPromoteToType<i32>>>,
CCIfInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16, bf16, v2bf16] , CCAssignToReg<
!foreach(i, !range(0, 30), !cast<Register>("SGPR"#i)) // SGPR0-29
@@ -167,6 +174,8 @@ def CC_AMDGPU : CallingConv<[
]>;
def CC_AMDGPU_CS_CHAIN : CallingConv<[
+ CCIfType<[i16, f16, bf16], CCIfInRegOrNoRealTrue16<CCPromoteToType<i32>>>,
+
CCIfInReg<CCIfType<[f32, i32, f16, i16, v2i16, v2f16, bf16, v2bf16] , CCAssignToReg<
!foreach(i, !range(105), !cast<Register>("SGPR"#i))
>>>,
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 3868bb084307d9..8bb2b968a23127 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -2724,6 +2724,11 @@ SDValue SITargetLowering::convertABITypeToValueType(SelectionDAG &DAG,
DAG.getValueType(ValVT));
return DAG.getNode(ISD::TRUNCATE, SL, ValVT, Val);
case CCValAssign::AExt:
+ if (ValVT.isFloatingPoint()) {
+ Val = DAG.getNode(ISD::TRUNCATE, SL, ValVT.changeTypeToInteger(), Val);
+ return DAG.getNode(ISD::BITCAST, SL, ValVT, Val);
+ }
+
return DAG.getNode(ISD::TRUNCATE, SL, ValVT, Val);
default:
llvm_unreachable("Unknown loc info!");
@@ -2750,6 +2755,10 @@ SDValue SITargetLowering::lowerStackParameter(SelectionDAG &DAG,
// Create load nodes to retrieve arguments from the stack.
SDValue FIN = DAG.getFrameIndex(FI, MVT::i32);
+ MachinePointerInfo PtrInfo =
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI);
+ if (VA.getLocInfo() == CCValAssign::AExt)
+ return DAG.getLoad(VA.getValVT(), SL, Chain, FIN, PtrInfo);
// For NON_EXTLOAD, generic code in getLoad assert(ValVT == MemVT)
ISD::LoadExtType ExtType = ISD::NON_EXTLOAD;
@@ -2767,14 +2776,10 @@ SDValue SITargetLowering::lowerStackParameter(SelectionDAG &DAG,
case CCValAssign::ZExt:
ExtType = ISD::ZEXTLOAD;
break;
- case CCValAssign::AExt:
- ExtType = ISD::EXTLOAD;
- break;
}
- SDValue ArgValue = DAG.getExtLoad(
- ExtType, SL, VA.getLocVT(), Chain, FIN,
- MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI), MemVT);
+ SDValue ArgValue =
+ DAG.getExtLoad(ExtType, SL, VA.getLocVT(), Chain, FIN, PtrInfo, MemVT);
SDValue ConvertedVal = convertABITypeToValueType(DAG, ArgValue, VA, SL);
if (ConvertedVal == ArgValue)
@@ -3822,10 +3827,7 @@ SDValue SITargetLowering::LowerFormalArguments(
llvm_unreachable("Unexpected register class in LowerFormalArguments!");
Reg = MF.addLiveIn(Reg, RC);
- bool CopyAsI32 =
- VT.getSizeInBits() == 16 &&
- (RC == &AMDGPU::SGPR_32RegClass || !Subtarget->useRealTrue16Insts());
- SDValue Val = DAG.getCopyFromReg(Chain, DL, Reg, CopyAsI32 ? MVT::i32 : VT);
+ SDValue Val = DAG.getCopyFromReg(Chain, DL, Reg, VT);
if (Arg.Flags.isInReg() && RC == &AMDGPU::VGPR_32RegClass) {
// FIXME: Need to forward the chains created by `CopyFromReg`s, make sure
// they will read physical regs before any side effect instructions.
@@ -3835,9 +3837,6 @@ SDValue SITargetLowering::LowerFormalArguments(
ReadFirstLane, Val);
}
- if (CopyAsI32)
- Val = DAG.getBitcast(VT, DAG.getNode(ISD::TRUNCATE, DL, MVT::i16, Val));
-
if (Arg.Flags.isSRet()) {
// The return object should be reasonably addressable.
Val = annotateStackObjectPointer(Val, DAG, DL,
@@ -4554,6 +4553,11 @@ SDValue SITargetLowering::LowerCall(CallLoweringInfo &CLI,
Arg = DAG.getNode(ISD::SIGN_EXTEND, DL, VA.getLocVT(), Arg);
break;
case CCValAssign::AExt:
+ if (VA.getValVT().isFloatingPoint()) {
+ Arg = DAG.getNode(ISD::BITCAST, DL, VA.getValVT().changeTypeToInteger(),
+ Arg);
+ }
+
Arg = DAG.getNode(ISD::ANY_EXTEND, DL, VA.getLocVT(), Arg);
break;
case CCValAssign::FPExt:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/and.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/and.ll
index ddc1f68e8fbe01..c5961ad97c6171 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/and.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/and.ll
@@ -53,34 +53,19 @@ define i16 @v_and_i16(i16 %num, i16 %den) {
}
define amdgpu_ps zeroext i16 @s_and_i16_zeroext(i16 inreg zeroext %num, i16 inreg zeroext %den) {
-; GFX7-LABEL: s_and_i16_zeroext:
-; GFX7: ; %bb.0:
-; GFX7-NEXT: s_and_b32 s0, s0, s1
-; GFX7-NEXT: ; return to shader part epilog
-;
-; GFX8-LABEL: s_and_i16_zeroext:
-; GFX8: ; %bb.0:
-; GFX8-NEXT: s_and_b32 s0, s0, s1
-; GFX8-NEXT: s_and_b32 s0, s0, 0xffff
-; GFX8-NEXT: ; return to shader part epilog
-;
-; GFX9-LABEL: s_and_i16_zeroext:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_and_b32 s0, s0, s1
-; GFX9-NEXT: s_and_b32 s0, s0, 0xffff
-; GFX9-NEXT: ; return to shader part epilog
+; GCN-LABEL: s_and_i16_zeroext:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_and_b32 s0, s0, s1
+; GCN-NEXT: ; return to shader part epilog
;
; GFX10PLUS-LABEL: s_and_i16_zeroext:
; GFX10PLUS: ; %bb.0:
; GFX10PLUS-NEXT: s_and_b32 s0, s0, s1
-; GFX10PLUS-NEXT: s_and_b32 s0, s0, 0xffff
; GFX10PLUS-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: s_and_i16_zeroext:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_and_b32 s0, s0, s1
-; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT: s_and_b32 s0, s0, 0xffff
; GFX12-NEXT: ; return to shader part epilog
%result = and i16 %num, %den
ret i16 %result
@@ -113,34 +98,19 @@ define zeroext i16 @v_and_i16_zeroext(i16 zeroext %num, i16 zeroext %den) {
}
define amdgpu_ps signext i16 @s_and_i16_signext(i16 inreg signext %num, i16 inreg signext %den) {
-; GFX7-LABEL: s_and_i16_signext:
-; GFX7: ; %bb.0:
-; GFX7-NEXT: s_and_b32 s0, s0, s1
-; GFX7-NEXT: ; return to shader part epilog
-;
-; GFX8-LABEL: s_and_i16_signext:
-; GFX8: ; %bb.0:
-; GFX8-NEXT: s_and_b32 s0, s0, s1
-; GFX8-NEXT: s_sext_i32_i16 s0, s0
-; GFX8-NEXT: ; return to shader part epilog
-;
-; GFX9-LABEL: s_and_i16_signext:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_and_b32 s0, s0, s1
-; GFX9-NEXT: s_sext_i32_i16 s0, s0
-; GFX9-NEXT: ; return to shader part epilog
+; GCN-LABEL: s_and_i16_signext:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_and_b32 s0, s0, s1
+; GCN-NEXT: ; return to shader part epilog
;
; GFX10PLUS-LABEL: s_and_i16_signext:
; GFX10PLUS: ; %bb.0:
; GFX10PLUS-NEXT: s_and_b32 s0, s0, s1
-; GFX10PLUS-NEXT: s_sext_i32_i16 s0, s0
; GFX10PLUS-NEXT: ; return to shader part epilog
;
; GFX12-LABEL: s_and_i16_signext:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_and_b32 s0, s0, s1
-; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT: s_sext_i32_i16 s0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = and i16 %num, %den
ret i16 %result
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-implicit-args.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-implicit-args.ll
index b652272f4f100d..dff625727f31d1 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-implicit-args.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-implicit-args.ll
@@ -403,56 +403,39 @@ define void @test_func_call_external_void_func_v32i32([17 x i8]) #0 {
; GFX900-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
; GFX900-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
; GFX900-NEXT: [[COPY9:%[0-9]+]]:_(i32) = COPY $vgpr0
- ; GFX900-NEXT: [[TRUNC:%[0-9]+]]:_(i16) = G_TRUNC [[COPY9]](i32)
- ; GFX900-NEXT: [[TRUNC1:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC]](i16)
+ ; GFX900-NEXT: [[TRUNC:%[0-9]+]]:_(i8) = G_TRUNC [[COPY9]](i32)
; GFX900-NEXT: [[COPY10:%[0-9]+]]:_(i32) = COPY $vgpr1
- ; GFX900-NEXT: [[TRUNC2:%[0-9]+]]:_(i16) = G_TRUNC [[COPY10]](i32)
- ; GFX900-NEXT: [[TRUNC3:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC2]](i16)
+ ; GFX900-NEXT: [[TRUNC1:%[0-9]+]]:_(i8) = G_TRUNC [[COPY10]](i32)
; GFX900-NEXT: [[COPY11:%[0-9]+]]:_(i32) = COPY $vgpr2
- ; GFX900-NEXT: [[TRUNC4:%[0-9]+]]:_(i16) = G_TRUNC [[COPY11]](i32)
- ; GFX900-NEXT: [[TRUNC5:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC4]](i16)
+ ; GFX900-NEXT: [[TRUNC2:%[0-9]+]]:_(i8) = G_TRUNC [[COPY11]](i32)
; GFX900-NEXT: [[COPY12:%[0-9]+]]:_(i32) = COPY $vgpr3
- ; GFX900-NEXT: [[TRUNC6:%[0-9]+]]:_(i16) = G_TRUNC [[COPY12]](i32)
- ; GFX900-NEXT: [[TRUNC7:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC6]](i16)
+ ; GFX900-NEXT: [[TRUNC3:%[0-9]+]]:_(i8) = G_TRUNC [[COPY12]](i32)
; GFX900-NEXT: [[COPY13:%[0-9]+]]:_(i32) = COPY $vgpr4
- ; GFX900-NEXT: [[TRUNC8:%[0-9]+]]:_(i16) = G_TRUNC [[COPY13]](i32)
- ; GFX900-NEXT: [[TRUNC9:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC8]](i16)
+ ; GFX900-NEXT: [[TRUNC4:%[0-9]+]]:_(i8) = G_TRUNC [[COPY13]](i32)
; GFX900-NEXT: [[COPY14:%[0-9]+]]:_(i32) = COPY $vgpr5
- ; GFX900-NEXT: [[TRUNC10:%[0-9]+]]:_(i16) = G_TRUNC [[COPY14]](i32)
- ; GFX900-NEXT: [[TRUNC11:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC10]](i16)
+ ; GFX900-NEXT: [[TRUNC5:%[0-9]+]]:_(i8) = G_TRUNC [[COPY14]](i32)
; GFX900-NEXT: [[COPY15:%[0-9]+]]:_(i32) = COPY $vgpr6
- ; GFX900-NEXT: [[TRUNC12:%[0-9]+]]:_(i16) = G_TRUNC [[COPY15]](i32)
- ; GFX900-NEXT: [[TRUNC13:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC12]](i16)
+ ; GFX900-NEXT: [[TRUNC6:%[0-9]+]]:_(i8) = G_TRUNC [[COPY15]](i32)
; GFX900-NEXT: [[COPY16:%[0-9]+]]:_(i32) = COPY $vgpr7
- ; GFX900-NEXT: [[TRUNC14:%[0-9]+]]:_(i16) = G_TRUNC [[COPY16]](i32)
- ; GFX900-NEXT: [[TRUNC15:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC14]](i16)
+ ; GFX900-NEXT: [[TRUNC7:%[0-9]+]]:_(i8) = G_TRUNC [[COPY16]](i32)
; GFX900-NEXT: [[COPY17:%[0-9]+]]:_(i32) = COPY $vgpr8
- ; GFX900-NEXT: [[TRUNC16:%[0-9]+]]:_(i16) = G_TRUNC [[COPY17]](i32)
- ; GFX900-NEXT: [[TRUNC17:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC16]](i16)
+ ; GFX900-NEXT: [[TRUNC8:%[0-9]+]]:_(i8) = G_TRUNC [[COPY17]](i32)
; GFX900-NEXT: [[COPY18:%[0-9]+]]:_(i32) = COPY $vgpr9
- ; GFX900-NEXT: [[TRUNC18:%[0-9]+]]:_(i16) = G_TRUNC [[COPY18]](i32)
- ; GFX900-NEXT: [[TRUNC19:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC18]](i16)
+ ; GFX900-NEXT: [[TRUNC9:%[0-9]+]]:_(i8) = G_TRUNC [[COPY18]](i32)
; GFX900-NEXT: [[COPY19:%[0-9]+]]:_(i32) = COPY $vgpr10
- ; GFX900-NEXT: [[TRUNC20:%[0-9]+]]:_(i16) = G_TRUNC [[COPY19]](i32)
- ; GFX900-NEXT: [[TRUNC21:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC20]](i16)
+ ; GFX900-NEXT: [[TRUNC10:%[0-9]+]]:_(i8) = G_TRUNC [[COPY19]](i32)
; GFX900-NEXT: [[COPY20:%[0-9]+]]:_(i32) = COPY $vgpr11
- ; GFX900-NEXT: [[TRUNC22:%[0-9]+]]:_(i16) = G_TRUNC [[COPY20]](i32)
- ; GFX900-NEXT: [[TRUNC23:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC22]](i16)
+ ; GFX900-NEXT: [[TRUNC11:%[0-9]+]]:_(i8) = G_TRUNC [[COPY20]](i32)
; GFX900-NEXT: [[COPY21:%[0-9]+]]:_(i32) = COPY $vgpr12
- ; GFX900-NEXT: [[TRUNC24:%[0-9]+]]:_(i16) = G_TRUNC [[COPY21]](i32)
- ; GFX900-NEXT: [[TRUNC25:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC24]](i16)
+ ; GFX900-NEXT: [[TRUNC12:%[0-9]+]]:_(i8) = G_TRUNC [[COPY21]](i32)
; GFX900-NEXT: [[COPY22:%[0-9]+]]:_(i32) = COPY $vgpr13
- ; GFX900-NEXT: [[TRUNC26:%[0-9]+]]:_(i16) = G_TRUNC [[COPY22]](i32)
- ; GFX900-NEXT: [[TRUNC27:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC26]](i16)
+ ; GFX900-NEXT: [[TRUNC13:%[0-9]+]]:_(i8) = G_TRUNC [[COPY22]](i32)
; GFX900-NEXT: [[COPY23:%[0-9]+]]:_(i32) = COPY $vgpr14
- ; GFX900-NEXT: [[TRUNC28:%[0-9]+]]:_(i16) = G_TRUNC [[COPY23]](i32)
- ; GFX900-NEXT: [[TRUNC29:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC28]](i16)
+ ; GFX900-NEXT: [[TRUNC14:%[0-9]+]]:_(i8) = G_TRUNC [[COPY23]](i32)
; GFX900-NEXT: [[COPY24:%[0-9]+]]:_(i32) = COPY $vgpr15
- ; GFX900-NEXT: [[TRUNC30:%[0-9]+]]:_(i16) = G_TRUNC [[COPY24]](i32)
- ; GFX900-NEXT: [[TRUNC31:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC30]](i16)
+ ; GFX900-NEXT: [[TRUNC15:%[0-9]+]]:_(i8) = G_TRUNC [[COPY24]](i32)
; GFX900-NEXT: [[COPY25:%[0-9]+]]:_(i32) = COPY $vgpr16
- ; GFX900-NEXT: [[TRUNC32:%[0-9]+]]:_(i16) = G_TRUNC [[COPY25]](i32)
- ; GFX900-NEXT: [[TRUNC33:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC32]](i16)
+ ; GFX900-NEXT: [[TRUNC16:%[0-9]+]]:_(i8) = G_TRUNC [[COPY25]](i32)
; GFX900-NEXT: [[C:%[0-9]+]]:_(i32) = G_CONSTANT i32 0
; GFX900-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<32 x i32>) = G_BUILD_VECTOR [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32)
; GFX900-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $scc
@@ -531,56 +514,39 @@ define void @test_func_call_external_void_func_v32i32([17 x i8]) #0 {
; GFX908-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
; GFX908-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
; GFX908-NEXT: [[COPY9:%[0-9]+]]:_(i32) = COPY $vgpr0
- ; GFX908-NEXT: [[TRUNC:%[0-9]+]]:_(i16) = G_TRUNC [[COPY9]](i32)
- ; GFX908-NEXT: [[TRUNC1:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC]](i16)
+ ; GFX908-NEXT: [[TRUNC:%[0-9]+]]:_(i8) = G_TRUNC [[COPY9]](i32)
; GFX908-NEXT: [[COPY10:%[0-9]+]]:_(i32) = COPY $vgpr1
- ; GFX908-NEXT: [[TRUNC2:%[0-9]+]]:_(i16) = G_TRUNC [[COPY10]](i32)
- ; GFX908-NEXT: [[TRUNC3:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC2]](i16)
+ ; GFX908-NEXT: [[TRUNC1:%[0-9]+]]:_(i8) = G_TRUNC [[COPY10]](i32)
; GFX908-NEXT: [[COPY11:%[0-9]+]]:_(i32) = COPY $vgpr2
- ; GFX908-NEXT: [[TRUNC4:%[0-9]+]]:_(i16) = G_TRUNC [[COPY11]](i32)
- ; GFX908-NEXT: [[TRUNC5:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC4]](i16)
+ ; GFX908-NEXT: [[TRUNC2:%[0-9]+]]:_(i8) = G_TRUNC [[COPY11]](i32)
; GFX908-NEXT: [[COPY12:%[0-9]+]]:_(i32) = COPY $vgpr3
- ; GFX908-NEXT: [[TRUNC6:%[0-9]+]]:_(i16) = G_TRUNC [[COPY12]](i32)
- ; GFX908-NEXT: [[TRUNC7:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC6]](i16)
+ ; GFX908-NEXT: [[TRUNC3:%[0-9]+]]:_(i8) = G_TRUNC [[COPY12]](i32)
; GFX908-NEXT: [[COPY13:%[0-9]+]]:_(i32) = COPY $vgpr4
- ; GFX908-NEXT: [[TRUNC8:%[0-9]+]]:_(i16) = G_TRUNC [[COPY13]](i32)
- ; GFX908-NEXT: [[TRUNC9:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC8]](i16)
+ ; GFX908-NEXT: [[TRUNC4:%[0-9]+]]:_(i8) = G_TRUNC [[COPY13]](i32)
; GFX908-NEXT: [[COPY14:%[0-9]+]]:_(i32) = COPY $vgpr5
- ; GFX908-NEXT: [[TRUNC10:%[0-9]+]]:_(i16) = G_TRUNC [[COPY14]](i32)
- ; GFX908-NEXT: [[TRUNC11:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC10]](i16)
+ ; GFX908-NEXT: [[TRUNC5:%[0-9]+]]:_(i8) = G_TRUNC [[COPY14]](i32)
; GFX908-NEXT: [[COPY15:%[0-9]+]]:_(i32) = COPY $vgpr6
- ; GFX908-NEXT: [[TRUNC12:%[0-9]+]]:_(i16) = G_TRUNC [[COPY15]](i32)
- ; GFX908-NEXT: [[TRUNC13:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC12]](i16)
+ ; GFX908-NEXT: [[TRUNC6:%[0-9]+]]:_(i8) = G_TRUNC [[COPY15]](i32)
; GFX908-NEXT: [[COPY16:%[0-9]+]]:_(i32) = COPY $vgpr7
- ; GFX908-NEXT: [[TRUNC14:%[0-9]+]]:_(i16) = G_TRUNC [[COPY16]](i32)
- ; GFX908-NEXT: [[TRUNC15:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC14]](i16)
+ ; GFX908-NEXT: [[TRUNC7:%[0-9]+]]:_(i8) = G_TRUNC [[COPY16]](i32)
; GFX908-NEXT: [[COPY17:%[0-9]+]]:_(i32) = COPY $vgpr8
- ; GFX908-NEXT: [[TRUNC16:%[0-9]+]]:_(i16) = G_TRUNC [[COPY17]](i32)
- ; GFX908-NEXT: [[TRUNC17:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC16]](i16)
+ ; GFX908-NEXT: [[TRUNC8:%[0-9]+]]:_(i8) = G_TRUNC [[COPY17]](i32)
; GFX908-NEXT: [[COPY18:%[0-9]+]]:_(i32) = COPY $vgpr9
- ; GFX908-NEXT: [[TRUNC18:%[0-9]+]]:_(i16) = G_TRUNC [[COPY18]](i32)
- ; GFX908-NEXT: [[TRUNC19:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC18]](i16)
+ ; GFX908-NEXT: [[TRUNC9:%[0-9]+]]:_(i8) = G_TRUNC [[COPY18]](i32)
; GFX908-NEXT: [[COPY19:%[0-9]+]]:_(i32) = COPY $vgpr10
- ; GFX908-NEXT: [[TRUNC20:%[0-9]+]]:_(i16) = G_TRUNC [[COPY19]](i32)
- ; GFX908-NEXT: [[TRUNC21:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC20]](i16)
+ ; GFX908-NEXT: [[TRUNC10:%[0-9]+]]:_(i8) = G_TRUNC [[COPY19]](i32)
; GFX908-NEXT: [[COPY20:%[0-9]+]]:_(i32) = COPY $vgpr11
- ; GFX908-NEXT: [[TRUNC22:%[0-9]+]]:_(i16) = G_TRUNC [[COPY20]](i32)
- ; GFX908-NEXT: [[TRUNC23:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC22]](i16)
+ ; GFX908-NEXT: [[TRUNC11:%[0-9]+]]:_(i8) = G_TRUNC [[COPY20]](i32)
; GFX908-NEXT: [[COPY21:%[0-9]+]]:_(i32) = COPY $vgpr12
- ; GFX908-NEXT: [[TRUNC24:%[0-9]+]]:_(i16) = G_TRUNC [[COPY21]](i32)
- ; GFX908-NEXT: [[TRUNC25:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC24]](i16)
+ ; GFX908-NEXT: [[TRUNC12:%[0-9]+]]:_(i8) = G_TRUNC [[COPY21]](i32)
; GFX908-NEXT: [[COPY22:%[0-9]+]]:_(i32) = COPY $vgpr13
- ; GFX908-NEXT: [[TRUNC26:%[0-9]+]]:_(i16) = G_TRUNC [[COPY22]](i32)
- ; GFX908-NEXT: [[TRUNC27:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC26]](i16)
+ ; GFX908-NEXT: [[TRUNC13:%[0-9]+]]:_(i8) = G_TRUNC [[COPY22]](i32)
; GFX908-NEXT: [[COPY23:%[0-9]+]]:_(i32) = COPY $vgpr14
- ; GFX908-NEXT: [[TRUNC28:%[0-9]+]]:_(i16) = G_TRUNC [[COPY23]](i32)
- ; GFX908-NEXT: [[TRUNC29:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC28]](i16)
+ ; GFX908-NEXT: [[TRUNC14:%[0-9]+]]:_(i8) = G_TRUNC [[COPY23]](i32)
; GFX908-NEXT: [[COPY24:%[0-9]+]]:_(i32) = COPY $vgpr15
- ; GFX908-NEXT: [[TRUNC30:%[0-9]+]]:_(i16) = G_TRUNC [[COPY24]](i32)
- ; GFX908-NEXT: [[TRUNC31:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC30]](i16)
+ ; GFX908-NEXT: [[TRUNC15:%[0-9]+]]:_(i8) = G_TRUNC [[COPY24]](i32)
; GFX908-NEXT: [[COPY25:%[0-9]+]]:_(i32) = COPY $vgpr16
- ; GFX908-NEXT: [[TRUNC32:%[0-9]+]]:_(i16) = G_TRUNC [[COPY25]](i32)
- ; GFX908-NEXT: [[TRUNC33:%[0-9]+]]:_(i8) = G_TRUNC [[TRUNC32]](i16)
+ ; GFX908-NEXT: [[TRUNC16:%[0-9]+]]:_(i8) = G_TRUNC [[COPY25]](i32)
; GFX908-NEXT: [[C:%[0-9]+]]:_(i32) = G_CONSTANT i32 0
; GFX908-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<32 x i32>) = G_BUILD_VECTOR [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32), [[C]](i32)
; GFX908-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $scc
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call.ll
index be5158852f6957..7d4e31ef01ef65 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call.ll
@@ -3769,14 +3769,17 @@ define amdgpu_kernel void @test_call_external_void_func_v32i32_i8_i8_i16() #0 {
; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:_(i16) = G_ANYEXT [[LOAD2]](i8)
; CHECK-NEXT: [[C4:%[0-9...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/242694
More information about the llvm-branch-commits
mailing list