shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity

[Matrix] Handle -fuse-matrix-tile-size=0 as tiling disabled. (#188861)

Browse Source 
Treat -fuse-matrix-tile-size=0 as disabling tiling, as tile-size of 0
doesn't really make sense.

Fixes https://github.com/llvm/llvm-project/issues/185153

PR: https://github.com/llvm/llvm-project/pull/188861
This commit is contained in:
Florian Hahn 2026-03-27 10:14:20 +00:00 committed by GitHub
parent 6a994417d8
commit b164e7c610
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 105 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
View File

@ -2139,7 +2139,7 @@ public:
LowerMatrixMultiplyFused(CallInst *MatMul,
SmallPtrSetImpl<Instruction *> &FusedInsts,
SmallVector<IntrinsicInst *, 16> &LifetimeEnds) {
if (!FuseMatrix || !DT)
if (!FuseMatrix || !DT || TileSize == 0)
return;
assert(AA && LI && "Analyses should be available");

104
llvm/test/Transforms/LowerMatrixIntrinsics/multiply-fused-tile-size-0.ll Normal file
View File

@ -0,0 +1,104 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -passes=lower-matrix-intrinsics -fuse-matrix-tile-size=0 -matrix-allow-contract -force-fuse-matrix %s -S | FileCheck %s
; REQUIRES: aarch64-registered-target
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "aarch64-apple-ios"
define void @multiply_2x2(ptr noalias %A, ptr noalias %B, ptr noalias %C) {
; CHECK-LABEL: define void @multiply_2x2(
; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], ptr noalias [[C:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[COL_LOAD:%.*]] = load <2 x double>, ptr [[A]], align 8
; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, ptr [[A]], i64 2
; CHECK-NEXT: [[COL_LOAD1:%.*]] = load <2 x double>, ptr [[VEC_GEP]], align 8
; CHECK-NEXT: [[COL_LOAD2:%.*]] = load <2 x double>, ptr [[B]], align 8
; CHECK-NEXT: [[VEC_GEP3:%.*]] = getelementptr double, ptr [[B]], i64 2
; CHECK-NEXT: [[COL_LOAD4:%.*]] = load <2 x double>, ptr [[VEC_GEP3]], align 8
; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[COL_LOAD2]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <2 x double> poison, double [[TMP0]], i64 0
; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP1:%.*]] = fmul contract <2 x double> [[BLOCK]], [[SPLAT_SPLAT]]
; CHECK-NEXT: [[BLOCK5:%.*]] = shufflevector <2 x double> [[COL_LOAD1]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[COL_LOAD2]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT6:%.*]] = insertelement <2 x double> poison, double [[TMP2]], i64 0
; CHECK-NEXT: [[SPLAT_SPLAT7:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT6]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP3:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[BLOCK5]], <2 x double> [[SPLAT_SPLAT7]], <2 x double> [[TMP1]])
; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <2 x double> [[TMP3]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP4]], <2 x i32> <i32 2, i32 3>
; CHECK-NEXT: [[BLOCK8:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[COL_LOAD4]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT9:%.*]] = insertelement <2 x double> poison, double [[TMP6]], i64 0
; CHECK-NEXT: [[SPLAT_SPLAT10:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT9]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP7:%.*]] = fmul contract <2 x double> [[BLOCK8]], [[SPLAT_SPLAT10]]
; CHECK-NEXT: [[BLOCK11:%.*]] = shufflevector <2 x double> [[COL_LOAD1]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[COL_LOAD4]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT12:%.*]] = insertelement <2 x double> poison, double [[TMP8]], i64 0
; CHECK-NEXT: [[SPLAT_SPLAT13:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT12]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP9:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[BLOCK11]], <2 x double> [[SPLAT_SPLAT13]], <2 x double> [[TMP7]])
; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <2 x double> [[TMP9]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP10]], <2 x i32> <i32 2, i32 3>
; CHECK-NEXT: store <2 x double> [[TMP5]], ptr [[C]], align 8
; CHECK-NEXT: [[VEC_GEP14:%.*]] = getelementptr double, ptr [[C]], i64 2
; CHECK-NEXT: store <2 x double> [[TMP11]], ptr [[VEC_GEP14]], align 8
; CHECK-NEXT: ret void
;
entry:
%a = load <4 x double>, ptr %A, align 8
%b = load <4 x double>, ptr %B, align 8
%c = call <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
store <4 x double> %c, ptr %C, align 8
ret void
}
define void @multiply_2x2_may_alias(ptr %A, ptr %B, ptr %C) {
; CHECK-LABEL: define void @multiply_2x2_may_alias(
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], ptr [[C:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[COL_LOAD:%.*]] = load <2 x double>, ptr [[A]], align 8
; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, ptr [[A]], i64 2
; CHECK-NEXT: [[COL_LOAD1:%.*]] = load <2 x double>, ptr [[VEC_GEP]], align 8
; CHECK-NEXT: [[COL_LOAD2:%.*]] = load <2 x double>, ptr [[B]], align 8
; CHECK-NEXT: [[VEC_GEP3:%.*]] = getelementptr double, ptr [[B]], i64 2
; CHECK-NEXT: [[COL_LOAD4:%.*]] = load <2 x double>, ptr [[VEC_GEP3]], align 8
; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[COL_LOAD2]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <2 x double> poison, double [[TMP0]], i64 0
; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP1:%.*]] = fmul contract <2 x double> [[BLOCK]], [[SPLAT_SPLAT]]
; CHECK-NEXT: [[BLOCK5:%.*]] = shufflevector <2 x double> [[COL_LOAD1]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[COL_LOAD2]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT6:%.*]] = insertelement <2 x double> poison, double [[TMP2]], i64 0
; CHECK-NEXT: [[SPLAT_SPLAT7:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT6]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP3:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[BLOCK5]], <2 x double> [[SPLAT_SPLAT7]], <2 x double> [[TMP1]])
; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <2 x double> [[TMP3]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP4]], <2 x i32> <i32 2, i32 3>
; CHECK-NEXT: [[BLOCK8:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[COL_LOAD4]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT9:%.*]] = insertelement <2 x double> poison, double [[TMP6]], i64 0
; CHECK-NEXT: [[SPLAT_SPLAT10:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT9]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP7:%.*]] = fmul contract <2 x double> [[BLOCK8]], [[SPLAT_SPLAT10]]
; CHECK-NEXT: [[BLOCK11:%.*]] = shufflevector <2 x double> [[COL_LOAD1]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[COL_LOAD4]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT12:%.*]] = insertelement <2 x double> poison, double [[TMP8]], i64 0
; CHECK-NEXT: [[SPLAT_SPLAT13:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT12]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP9:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[BLOCK11]], <2 x double> [[SPLAT_SPLAT13]], <2 x double> [[TMP7]])
; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <2 x double> [[TMP9]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP10]], <2 x i32> <i32 2, i32 3>
; CHECK-NEXT: store <2 x double> [[TMP5]], ptr [[C]], align 8
; CHECK-NEXT: [[VEC_GEP14:%.*]] = getelementptr double, ptr [[C]], i64 2
; CHECK-NEXT: store <2 x double> [[TMP11]], ptr [[VEC_GEP14]], align 8
; CHECK-NEXT: ret void
;
entry:
%a = load <4 x double>, ptr %A, align 8
%b = load <4 x double>, ptr %B, align 8
%c = call <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
store <4 x double> %c, ptr %C, align 8
ret void
}
declare <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double>, <4 x double>, i32 immarg, i32 immarg, i32 immarg)