[SPIR-V] Add vector type support for non-standard integers in G_TRUNC op - #198213
Merged
Merged
Conversation
|
@llvm/pr-subscribers-backend-spir-v Author: Arseniy Obolenskiy (aobolensk) ChangesFull diff: https://github.com/llvm/llvm-project/pull/198213.diff 2 Files Affected:
diff --git a/llvm/lib/Target/SPIRV/SPIRVPreLegalizer.cpp b/llvm/lib/Target/SPIRV/SPIRVPreLegalizer.cpp
index 75cc7133f1766..14fc8b2fc4001 100644
--- a/llvm/lib/Target/SPIRV/SPIRVPreLegalizer.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVPreLegalizer.cpp
@@ -514,30 +514,44 @@ generateAssignInstrs(MachineFunction &MF, SPIRVGlobalRegistry *GR,
Register DstReg = MI.getOperand(0).getReg();
Register SrcReg = MI.getOperand(1).getReg();
- // TODO: handle vector types.
- if (!MRI.getType(DstReg).isScalar()) {
- assert(!MRI.getType(SrcReg).isScalar());
- continue;
- }
-
- unsigned OriginalDstWidth = MRI.getType(DstReg).getScalarSizeInBits();
- unsigned OriginalSrcWidth = MRI.getType(SrcReg).getScalarSizeInBits();
+ LLT DstTy = MRI.getType(DstReg);
+ LLT SrcTy = MRI.getType(SrcReg);
+ assert((DstTy.isScalar() || DstTy.isVector()) &&
+ (SrcTy.isScalar() || SrcTy.isVector()) &&
+ "Expected scalar or vector G_TRUNC types");
+ assert(DstTy.isVector() == SrcTy.isVector() &&
+ "Expected matching scalar/vector G_TRUNC types");
+ assert((!DstTy.isVector() ||
+ DstTy.getElementCount() == SrcTy.getElementCount()) &&
+ "Expected equal vector element counts");
+
+ unsigned OriginalDstWidth = DstTy.getScalarSizeInBits();
+ unsigned OriginalSrcWidth = SrcTy.getScalarSizeInBits();
unsigned NewDstWidth = widenBitWidthToNextPow2(OriginalDstWidth);
unsigned NewSrcWidth = widenBitWidthToNextPow2(OriginalSrcWidth);
+ LLT NewDstTy = DstTy.changeElementSize(NewDstWidth);
+ LLT NewSrcTy = SrcTy.changeElementSize(NewSrcWidth);
- // No Dst width change means no truncation semantics change.
- if (OriginalDstWidth == NewDstWidth)
+ // No Dst width change means no truncation semantics change, but the
+ // source still needs a legal type.
+ if (OriginalDstWidth == NewDstWidth) {
+ MRI.setType(SrcReg, NewSrcTy);
continue;
+ }
- MRI.setType(SrcReg, LLT::scalar(NewSrcWidth));
- MRI.setType(DstReg, LLT::scalar(NewDstWidth));
+ MRI.setType(SrcReg, NewSrcTy);
+ MRI.setType(DstReg, NewDstTy);
MIB.setInsertPt(MBB, MI.getIterator());
APInt Mask = APInt::getLowBitsSet(NewSrcWidth, OriginalDstWidth);
- auto MaskReg = MIB.buildConstant(LLT::scalar(NewSrcWidth), Mask);
- Register MaskedReg =
- MRI.createGenericVirtualRegister(LLT::scalar(NewSrcWidth));
+ MachineInstrBuilder MaskReg =
+ DstTy.isVector()
+ ? MIB.buildBuildVectorConstant(
+ NewSrcTy,
+ SmallVector<APInt, 4>(DstTy.getNumElements(), Mask))
+ : MIB.buildConstant(NewSrcTy, Mask);
+ Register MaskedReg = MRI.createGenericVirtualRegister(NewSrcTy);
MIB.buildAnd(MaskedReg, SrcReg, MaskReg);
if (NewSrcWidth == NewDstWidth) {
diff --git a/llvm/test/CodeGen/SPIRV/trunc-nonstd-bitwidth.ll b/llvm/test/CodeGen/SPIRV/trunc-nonstd-bitwidth.ll
index 22af0c3b81ee0..e3b2f48ceead3 100644
--- a/llvm/test/CodeGen/SPIRV/trunc-nonstd-bitwidth.ll
+++ b/llvm/test/CodeGen/SPIRV/trunc-nonstd-bitwidth.ll
@@ -13,8 +13,26 @@
; CHECK-EXT-DAG: %[[#Int40:]] = OpTypeInt 40 0
; CHECK-EXT-DAG: %[[#Int50:]] = OpTypeInt 50 0
; CHECK-EXT-DAG: %[[#Int24:]] = OpTypeInt 24 0
+; CHECK-EXT-DAG: %[[#ExtInt32:]] = OpTypeInt 32 0
; CHECK-NOEXT-DAG: %[[#Int64:]] = OpTypeInt 64 0
; CHECK-NOEXT-DAG: %[[#Int32:]] = OpTypeInt 32 0
+; CHECK-EXT-DAG: %[[#Vec2Int40:]] = OpTypeVector %[[#Int40]] 2
+; CHECK-EXT-DAG: %[[#Vec2Int50:]] = OpTypeVector %[[#Int50]] 2
+; CHECK-EXT-DAG: %[[#Vec2ExtInt32:]] = OpTypeVector %[[#ExtInt32]] 2
+; CHECK-EXT-DAG: %[[#Vec3Int50:]] = OpTypeVector %[[#Int50]] 3
+; CHECK-EXT-DAG: %[[#Vec3Int24:]] = OpTypeVector %[[#Int24]] 3
+; CHECK-EXT-DAG: %[[#Vec4Int24:]] = OpTypeVector %[[#Int24]] 4
+; CHECK-NOEXT-DAG: %[[#Vec2Int64:]] = OpTypeVector %[[#Int64]] 2
+; CHECK-NOEXT-DAG: %[[#Vec2Int32:]] = OpTypeVector %[[#Int32]] 2
+; CHECK-NOEXT-DAG: %[[#Vec3Int64:]] = OpTypeVector %[[#Int64]] 3
+; CHECK-NOEXT-DAG: %[[#Vec4Int64:]] = OpTypeVector %[[#Int64]] 4
+; CHECK-NOEXT-DAG: %[[#Vec3Int32:]] = OpTypeVector %[[#Int32]] 3
+; CHECK-NOEXT-DAG: %[[#Vec4Int32:]] = OpTypeVector %[[#Int32]] 4
+; CHECK-NOEXT-DAG: %[[#Mask40:]] = OpConstant %[[#Int64]] 1099511627775
+; CHECK-NOEXT-DAG: %[[#Mask24:]] = OpConstant %[[#Int64]] 16777215
+; CHECK-NOEXT-DAG: %[[#Mask40Vec2:]] = OpConstantComposite %[[#Vec2Int64]] %[[#Mask40]] %[[#Mask40]]
+; CHECK-NOEXT-DAG: %[[#Mask24Vec3:]] = OpConstantComposite %[[#Vec3Int64]] %[[#Mask24]] %[[#Mask24]] %[[#Mask24]]
+; CHECK-NOEXT-DAG: %[[#Mask24Vec4:]] = OpConstantComposite %[[#Vec4Int64]] %[[#Mask24]] %[[#Mask24]] %[[#Mask24]] %[[#Mask24]]
; Test i64 -> i40: both widen to i64
@@ -60,3 +78,62 @@ define spir_kernel void @trunc_i64_to_i24(ptr addrspace(1) %arg, i64 %val) {
store i24 %tr, ptr addrspace(1) %arg
ret void
}
+
+; Test <2 x i64> -> <2 x i40>: both widen to <2 x i64>
+; CHECK: OpFunction
+; CHECK: %[[#T4Arg:]] = OpFunctionParameter
+; CHECK: %[[#T4Val:]] = OpFunctionParameter
+; CHECK-EXT: %[[#T4Tr:]] = OpUConvert %[[#Vec2Int40]] %[[#T4Val]]
+; CHECK-EXT: OpStore %[[#T4Arg]] %[[#T4Tr]]
+; CHECK-NOEXT: %[[#T4And:]] = OpBitwiseAnd %[[#Vec2Int64]] %[[#T4Val]] %[[#Mask40Vec2]]
+; CHECK-NOEXT: OpStore %[[#T4Arg]] %[[#T4And]]
+define spir_kernel void @trunc_v2i64_to_v2i40(ptr addrspace(1) %arg, <2 x i64> %val) {
+ %tr = trunc <2 x i64> %val to <2 x i40>
+ store <2 x i40> %tr, ptr addrspace(1) %arg
+ ret void
+}
+
+; Test <3 x i50> -> <3 x i24>: src widens to <3 x i64>, dst widens to <3 x i32>
+; CHECK: OpFunction
+; CHECK: %[[#T5Arg:]] = OpFunctionParameter
+; CHECK: %[[#T5Val:]] = OpFunctionParameter
+; CHECK-EXT: %[[#T5Tr:]] = OpUConvert %[[#Vec3Int24]] %[[#T5Val]]
+; CHECK-EXT: OpStore %[[#T5Arg]] %[[#T5Tr]]
+; CHECK-NOEXT: %[[#T5And:]] = OpBitwiseAnd %[[#Vec3Int64]] %[[#T5Val]] %[[#Mask24Vec3]]
+; CHECK-NOEXT: %[[#T5Conv:]] = OpUConvert %[[#Vec3Int32]] %[[#T5And]]
+; CHECK-NOEXT: OpStore %[[#T5Arg]] %[[#T5Conv]]
+define spir_kernel void @trunc_v3i50_to_v3i24(ptr addrspace(1) %arg, <3 x i50> %val) {
+ %tr = trunc <3 x i50> %val to <3 x i24>
+ store <3 x i24> %tr, ptr addrspace(1) %arg
+ ret void
+}
+
+; Test <4 x i64> -> <4 x i24>: src stays <4 x i64>, dst widens to <4 x i32>
+; CHECK: OpFunction
+; CHECK: %[[#T6Arg:]] = OpFunctionParameter
+; CHECK: %[[#T6Val:]] = OpFunctionParameter
+; CHECK-EXT: %[[#T6Tr:]] = OpUConvert %[[#Vec4Int24]] %[[#T6Val]]
+; CHECK-EXT: OpStore %[[#T6Arg]] %[[#T6Tr]]
+; CHECK-NOEXT: %[[#T6And:]] = OpBitwiseAnd %[[#Vec4Int64]] %[[#T6Val]] %[[#Mask24Vec4]]
+; CHECK-NOEXT: %[[#T6Conv:]] = OpUConvert %[[#Vec4Int32]] %[[#T6And]]
+; CHECK-NOEXT: OpStore %[[#T6Arg]] %[[#T6Conv]]
+define spir_kernel void @trunc_v4i64_to_v4i24(ptr addrspace(1) %arg, <4 x i64> %val) {
+ %tr = trunc <4 x i64> %val to <4 x i24>
+ store <4 x i24> %tr, ptr addrspace(1) %arg
+ ret void
+}
+
+; Test <2 x i50> -> <2 x i32>: dst width is already legal, no mask needed
+; CHECK: OpFunction
+; CHECK: %[[#T7Arg:]] = OpFunctionParameter
+; CHECK: %[[#T7Val:]] = OpFunctionParameter
+; CHECK-EXT: %[[#T7Tr:]] = OpUConvert %[[#Vec2ExtInt32]] %[[#T7Val]]
+; CHECK-EXT: OpStore %[[#T7Arg]] %[[#T7Tr]]
+; CHECK-NOEXT-NOT: OpBitwiseAnd
+; CHECK-NOEXT: %[[#T7Conv:]] = OpUConvert %[[#Vec2Int32]] %[[#T7Val]]
+; CHECK-NOEXT: OpStore %[[#T7Arg]] %[[#T7Conv]]
+define spir_kernel void @trunc_v2i50_to_v2i32(ptr addrspace(1) %arg, <2 x i50> %val) {
+ %tr = trunc <2 x i50> %val to <2 x i32>
+ store <2 x i32> %tr, ptr addrspace(1) %arg
+ ret void
+}
|
MrSidims
approved these changes
Jun 7, 2026
MrSidims
left a comment
MrSidims
left a comment
Contributor
There was a problem hiding this comment.
Thanks!
Q: This fixes G_TRUNC specifically, but widenScalarType still early-returns on vectors, so other vector ops on non-standard int element types remain un-widened. Is G_TRUNC the only op that currently reaches here with such types, or is this the first of several follow-ups? Worth a note in the description.
MrSidims
requested review from
idubinov,
jmmartinez,
michalpaszkowski,
sarnex and
vmaksimo
June 7, 2026 19:44
Contributor
Author
So far, |
carlobertolli
pushed a commit
to carlobertolli/llvm-project
that referenced
this pull request
Jun 11, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
G_TRUNCis the special case for now that require these changes because it is the only one that depend on the original destination