[LMI] Support non-power-of-2 types for the matmul remainder

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

@@ -97,6 +97,12 @@ static cl::opt<MatrixLayoutTy> MatrixLayout(
 static cl::opt<bool> PrintAfterTransposeOpt("matrix-print-after-transpose-opt",
                                             cl::init(false));
 
+static cl::opt<unsigned> SplitMatmulRemainderOverThreshold(
+    "matrix-split-matmul-remainder-over-threshold", cl::Hidden,
+    cl::desc("Illegal remainder vectors over this size in bits should be split "
+             "in the inner loop of matmul"),
+    cl::init(0));
+
 /// Helper function to either return Scope, if it is a subprogram or the
 /// attached subprogram for a local scope.
 static DISubprogram *getSubprogram(DIScope *Scope) {
@@ -1720,6 +1726,31 @@ class LowerMatrixIntrinsics {
     ToRemove.push_back(MatMul);
   }
 
+  /// Given \p Remainder iterations of the the matmul inner loop,
+  /// potentially lower \p Blocksize that is used for the underlying
+  /// vector.
+  unsigned capBlockSize(unsigned BlockSize, unsigned Remainder, Type *EltType) {
+    if (BlockSize <= Remainder)
+      return BlockSize;
+
+    // If the remainder is also a legal type just use it.
+    auto *VecTy = FixedVectorType::get(EltType, Remainder);
+    if (TTI.isTypeLegal(VecTy))
+      return Remainder;
+
+    // Similarly, if the vector is small enough that we don't want
+    // to split further.
+    if (VecTy->getPrimitiveSizeInBits() <= SplitMatmulRemainderOverThreshold)
+      return Remainder;
+
+    // Gradually lower the vectorization factor to cover the
+    // remainder.
+    do {
+      BlockSize /= 2;
+    } while (BlockSize > Remainder);
+    return BlockSize;
+  }
+
   /// Compute \p Result += \p A * \p B for input matrices with left-associating
   /// addition.
   ///
@@ -1757,10 +1788,8 @@ class LowerMatrixIntrinsics {
         bool isSumZero = isa<ConstantAggregateZero>(Result.getColumn(J));
 
         for (unsigned I = 0; I < R; I += BlockSize) {
-          // Gradually lower the vectorization factor to cover the remainder.
-          while (I + BlockSize > R)
-            BlockSize /= 2;
-
+          // Lower block size to make sure we stay within bounds.
+          BlockSize = capBlockSize(BlockSize, R - I, Result.getElementType());
           Value *Sum = IsTiled ? Result.extractVector(I, J, BlockSize, Builder)
                                : nullptr;
           for (unsigned K = 0; K < M; ++K) {
@@ -1785,9 +1814,8 @@ class LowerMatrixIntrinsics {
         unsigned BlockSize = VF;
         bool isSumZero = isa<ConstantAggregateZero>(Result.getRow(I));
         for (unsigned J = 0; J < C; J += BlockSize) {
-          // Gradually lower the vectorization factor to cover the remainder.
-          while (J + BlockSize > C)
-            BlockSize /= 2;
+          // Lower the vectorization factor to cover the remainder.
+          BlockSize = capBlockSize(BlockSize, C - J, Result.getElementType());
 
           Value *Sum = nullptr;
           for (unsigned K = 0; K < M; ++K) {

llvm/test/Transforms/LowerMatrixIntrinsics/multiply-remainder-rm.ll

@@ -0,0 +1,96 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes='lower-matrix-intrinsics'                                                  -matrix-default-layout=row-major -S < %s | FileCheck --check-prefix=SPLIT_REMAINDER %s
+; RUN: opt -passes='lower-matrix-intrinsics' -matrix-split-matmul-remainder-over-threshold=96 -matrix-default-layout=row-major -S < %s | FileCheck --check-prefix=NO_SPLIT_REMAINDER %s
+; RUN: opt -passes='lower-matrix-intrinsics' -matrix-split-matmul-remainder-over-threshold=64 -matrix-default-layout=row-major -S < %s | FileCheck --check-prefix=SPLIT_REMAINDER %s
+
+; REQUIRES: aarch64-registered-target
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:8:32:64-S128"
+target triple = "aarch64-apple-ios"
+
+define void @matmul(ptr %a, ptr %b, ptr %c) {
+; SPLIT_REMAINDER-LABEL: define void @matmul(
+; SPLIT_REMAINDER-SAME: ptr [[A:%.*]], ptr [[B:%.*]], ptr [[C:%.*]]) {
+; SPLIT_REMAINDER-NEXT:    [[COL_LOAD:%.*]] = load <3 x float>, ptr [[A]], align 4
+; SPLIT_REMAINDER-NEXT:    [[COL_LOAD1:%.*]] = load <3 x float>, ptr [[B]], align 4
+; SPLIT_REMAINDER-NEXT:    [[VEC_GEP:%.*]] = getelementptr float, ptr [[B]], i64 3
+; SPLIT_REMAINDER-NEXT:    [[COL_LOAD2:%.*]] = load <3 x float>, ptr [[VEC_GEP]], align 4
+; SPLIT_REMAINDER-NEXT:    [[VEC_GEP3:%.*]] = getelementptr float, ptr [[B]], i64 6
+; SPLIT_REMAINDER-NEXT:    [[COL_LOAD4:%.*]] = load <3 x float>, ptr [[VEC_GEP3]], align 4
+; SPLIT_REMAINDER-NEXT:    [[BLOCK:%.*]] = shufflevector <3 x float> [[COL_LOAD1]], <3 x float> poison, <2 x i32> <i32 0, i32 1>
+; SPLIT_REMAINDER-NEXT:    [[TMP1:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 0
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[TMP1]], i64 0
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x float> [[SPLAT_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
+; SPLIT_REMAINDER-NEXT:    [[TMP2:%.*]] = fmul <2 x float> [[SPLAT_SPLAT]], [[BLOCK]]
+; SPLIT_REMAINDER-NEXT:    [[BLOCK5:%.*]] = shufflevector <3 x float> [[COL_LOAD2]], <3 x float> poison, <2 x i32> <i32 0, i32 1>
+; SPLIT_REMAINDER-NEXT:    [[TMP3:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 1
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT6:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT7:%.*]] = shufflevector <2 x float> [[SPLAT_SPLATINSERT6]], <2 x float> poison, <2 x i32> zeroinitializer
+; SPLIT_REMAINDER-NEXT:    [[TMP4:%.*]] = fmul <2 x float> [[SPLAT_SPLAT7]], [[BLOCK5]]
+; SPLIT_REMAINDER-NEXT:    [[TMP5:%.*]] = fadd <2 x float> [[TMP2]], [[TMP4]]
+; SPLIT_REMAINDER-NEXT:    [[BLOCK8:%.*]] = shufflevector <3 x float> [[COL_LOAD4]], <3 x float> poison, <2 x i32> <i32 0, i32 1>
+; SPLIT_REMAINDER-NEXT:    [[TMP6:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 2
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT9:%.*]] = insertelement <2 x float> poison, float [[TMP6]], i64 0
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT10:%.*]] = shufflevector <2 x float> [[SPLAT_SPLATINSERT9]], <2 x float> poison, <2 x i32> zeroinitializer
+; SPLIT_REMAINDER-NEXT:    [[TMP7:%.*]] = fmul <2 x float> [[SPLAT_SPLAT10]], [[BLOCK8]]
+; SPLIT_REMAINDER-NEXT:    [[TMP8:%.*]] = fadd <2 x float> [[TMP5]], [[TMP7]]
+; SPLIT_REMAINDER-NEXT:    [[TMP9:%.*]] = shufflevector <2 x float> [[TMP8]], <2 x float> poison, <3 x i32> <i32 0, i32 1, i32 poison>
+; SPLIT_REMAINDER-NEXT:    [[TMP10:%.*]] = shufflevector <3 x float> poison, <3 x float> [[TMP9]], <3 x i32> <i32 3, i32 4, i32 2>
+; SPLIT_REMAINDER-NEXT:    [[BLOCK11:%.*]] = shufflevector <3 x float> [[COL_LOAD1]], <3 x float> poison, <1 x i32> <i32 2>
+; SPLIT_REMAINDER-NEXT:    [[TMP11:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 0
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT12:%.*]] = insertelement <1 x float> poison, float [[TMP11]], i64 0
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT13:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT12]], <1 x float> poison, <1 x i32> zeroinitializer
+; SPLIT_REMAINDER-NEXT:    [[TMP12:%.*]] = fmul <1 x float> [[SPLAT_SPLAT13]], [[BLOCK11]]
+; SPLIT_REMAINDER-NEXT:    [[BLOCK14:%.*]] = shufflevector <3 x float> [[COL_LOAD2]], <3 x float> poison, <1 x i32> <i32 2>
+; SPLIT_REMAINDER-NEXT:    [[TMP13:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 1
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT15:%.*]] = insertelement <1 x float> poison, float [[TMP13]], i64 0
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT16:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT15]], <1 x float> poison, <1 x i32> zeroinitializer
+; SPLIT_REMAINDER-NEXT:    [[TMP14:%.*]] = fmul <1 x float> [[SPLAT_SPLAT16]], [[BLOCK14]]
+; SPLIT_REMAINDER-NEXT:    [[TMP15:%.*]] = fadd <1 x float> [[TMP12]], [[TMP14]]
+; SPLIT_REMAINDER-NEXT:    [[BLOCK17:%.*]] = shufflevector <3 x float> [[COL_LOAD4]], <3 x float> poison, <1 x i32> <i32 2>
+; SPLIT_REMAINDER-NEXT:    [[TMP16:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 2
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT18:%.*]] = insertelement <1 x float> poison, float [[TMP16]], i64 0
+; SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT19:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT18]], <1 x float> poison, <1 x i32> zeroinitializer
+; SPLIT_REMAINDER-NEXT:    [[TMP17:%.*]] = fmul <1 x float> [[SPLAT_SPLAT19]], [[BLOCK17]]
+; SPLIT_REMAINDER-NEXT:    [[TMP18:%.*]] = fadd <1 x float> [[TMP15]], [[TMP17]]
+; SPLIT_REMAINDER-NEXT:    [[TMP19:%.*]] = shufflevector <1 x float> [[TMP18]], <1 x float> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; SPLIT_REMAINDER-NEXT:    [[TMP20:%.*]] = shufflevector <3 x float> [[TMP10]], <3 x float> [[TMP19]], <3 x i32> <i32 0, i32 1, i32 3>
+; SPLIT_REMAINDER-NEXT:    store <3 x float> [[TMP20]], ptr [[C]], align 4
+; SPLIT_REMAINDER-NEXT:    ret void
+;
+; NO_SPLIT_REMAINDER-LABEL: define void @matmul(
+; NO_SPLIT_REMAINDER-SAME: ptr [[A:%.*]], ptr [[B:%.*]], ptr [[C:%.*]]) {
+; NO_SPLIT_REMAINDER-NEXT:    [[COL_LOAD:%.*]] = load <3 x float>, ptr [[A]], align 4
+; NO_SPLIT_REMAINDER-NEXT:    [[COL_LOAD1:%.*]] = load <3 x float>, ptr [[B]], align 4
+; NO_SPLIT_REMAINDER-NEXT:    [[VEC_GEP:%.*]] = getelementptr float, ptr [[B]], i64 3
+; NO_SPLIT_REMAINDER-NEXT:    [[COL_LOAD2:%.*]] = load <3 x float>, ptr [[VEC_GEP]], align 4
+; NO_SPLIT_REMAINDER-NEXT:    [[VEC_GEP3:%.*]] = getelementptr float, ptr [[B]], i64 6
+; NO_SPLIT_REMAINDER-NEXT:    [[COL_LOAD4:%.*]] = load <3 x float>, ptr [[VEC_GEP3]], align 4
+; NO_SPLIT_REMAINDER-NEXT:    [[BLOCK:%.*]] = shufflevector <3 x float> [[COL_LOAD1]], <3 x float> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP1:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 0
+; NO_SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <3 x float> poison, float [[TMP1]], i64 0
+; NO_SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <3 x float> [[SPLAT_SPLATINSERT]], <3 x float> poison, <3 x i32> zeroinitializer
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP2:%.*]] = fmul <3 x float> [[SPLAT_SPLAT]], [[BLOCK]]
+; NO_SPLIT_REMAINDER-NEXT:    [[BLOCK5:%.*]] = shufflevector <3 x float> [[COL_LOAD2]], <3 x float> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP3:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 1
+; NO_SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT6:%.*]] = insertelement <3 x float> poison, float [[TMP3]], i64 0
+; NO_SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT7:%.*]] = shufflevector <3 x float> [[SPLAT_SPLATINSERT6]], <3 x float> poison, <3 x i32> zeroinitializer
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP4:%.*]] = fmul <3 x float> [[SPLAT_SPLAT7]], [[BLOCK5]]
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP5:%.*]] = fadd <3 x float> [[TMP2]], [[TMP4]]
+; NO_SPLIT_REMAINDER-NEXT:    [[BLOCK8:%.*]] = shufflevector <3 x float> [[COL_LOAD4]], <3 x float> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP6:%.*]] = extractelement <3 x float> [[COL_LOAD]], i64 2
+; NO_SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLATINSERT9:%.*]] = insertelement <3 x float> poison, float [[TMP6]], i64 0
+; NO_SPLIT_REMAINDER-NEXT:    [[SPLAT_SPLAT10:%.*]] = shufflevector <3 x float> [[SPLAT_SPLATINSERT9]], <3 x float> poison, <3 x i32> zeroinitializer
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP7:%.*]] = fmul <3 x float> [[SPLAT_SPLAT10]], [[BLOCK8]]
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP8:%.*]] = fadd <3 x float> [[TMP5]], [[TMP7]]
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP9:%.*]] = shufflevector <3 x float> [[TMP8]], <3 x float> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; NO_SPLIT_REMAINDER-NEXT:    [[TMP10:%.*]] = shufflevector <3 x float> poison, <3 x float> [[TMP9]], <3 x i32> <i32 3, i32 4, i32 5>
+; NO_SPLIT_REMAINDER-NEXT:    store <3 x float> [[TMP10]], ptr [[C]], align 4
+; NO_SPLIT_REMAINDER-NEXT:    ret void
+;
+  %a_load = load <3 x float>, ptr %a, align 4
+  %b_load = load <9 x float>, ptr %b, align 4
+  %matmul = tail call <3 x float> @llvm.matrix.multiply.v3f32.v9f32.v3f32(<3 x float> %a_load, <9 x float> %b_load, i32 1, i32 3, i32 3)
+  store <3 x float> %matmul, ptr %c, align 4
+  ret void
+}