[PATCH] D144455: [RISCV] Enable ordered reduction.

[Mel Chen via Phabricator via llvm-commits]

Mel-Chen updated this revision to Diff 499076.
Mel-Chen added a comment.

Create another NFC revision.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D144455/new/

https://reviews.llvm.org/D144455

Files:
  llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
  llvm/test/Transforms/LoopVectorize/RISCV/ordered-reduction.ll


Index: llvm/test/Transforms/LoopVectorize/RISCV/ordered-reduction.ll
===================================================================
--- llvm/test/Transforms/LoopVectorize/RISCV/ordered-reduction.ll
+++ llvm/test/Transforms/LoopVectorize/RISCV/ordered-reduction.ll
@@ -1,11 +1,50 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt -passes=loop-vectorize -hints-allow-reordering=false -S < %s | FileCheck %s --check-prefix=CHECK-NOT-VECTORIZED
+; RUN: opt -passes=loop-vectorize -hints-allow-reordering=false -S < %s | FileCheck %s --check-prefix=CHECK-ORDERED
 ; RUN: opt -passes=loop-vectorize -S < %s | FileCheck %s --check-prefix=CHECK-UNORDERED
 ; RUN: opt -passes=loop-vectorize -force-ordered-reductions=false -hints-allow-reordering=false -S < %s | FileCheck %s --check-prefix=CHECK-NOT-VECTORIZED
 
 target triple = "riscv64"
 
 define float @fadd(ptr noalias nocapture readonly %a, i64 %n) #0 {
+; CHECK-ORDERED-LABEL: @fadd(
+; CHECK-ORDERED-NEXT:  entry:
+; CHECK-ORDERED-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
+; CHECK-ORDERED-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK-ORDERED:       vector.ph:
+; CHECK-ORDERED-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
+; CHECK-ORDERED-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
+; CHECK-ORDERED-NEXT:    br label [[VECTOR_BODY:%.*]]
+; CHECK-ORDERED:       vector.body:
+; CHECK-ORDERED-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-ORDERED-NEXT:    [[VEC_PHI:%.*]] = phi float [ 0.000000e+00, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ]
+; CHECK-ORDERED-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-ORDERED-NEXT:    [[TMP1:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[TMP0]]
+; CHECK-ORDERED-NEXT:    [[TMP2:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i32 0
+; CHECK-ORDERED-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP2]], align 4
+; CHECK-ORDERED-NEXT:    [[TMP3]] = call float @llvm.vector.reduce.fadd.v4f32(float [[VEC_PHI]], <4 x float> [[WIDE_LOAD]])
+; CHECK-ORDERED-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-ORDERED-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-ORDERED-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-ORDERED:       middle.block:
+; CHECK-ORDERED-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
+; CHECK-ORDERED-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
+; CHECK-ORDERED:       scalar.ph:
+; CHECK-ORDERED-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-ORDERED-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
+; CHECK-ORDERED-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK-ORDERED:       for.body:
+; CHECK-ORDERED-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
+; CHECK-ORDERED-NEXT:    [[SUM_07:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
+; CHECK-ORDERED-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]]
+; CHECK-ORDERED-NEXT:    [[TMP5:%.*]] = load float, ptr [[ARRAYIDX]], align 4
+; CHECK-ORDERED-NEXT:    [[ADD]] = fadd float [[TMP5]], [[SUM_07]]
+; CHECK-ORDERED-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-ORDERED-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
+; CHECK-ORDERED-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-ORDERED:       for.end:
+; CHECK-ORDERED-NEXT:    [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], [[FOR_BODY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
+; CHECK-ORDERED-NEXT:    ret float [[ADD_LCSSA]]
+;
 ; CHECK-UNORDERED-LABEL: @fadd(
 ; CHECK-UNORDERED-NEXT:  entry:
 ; CHECK-UNORDERED-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
Index: llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
===================================================================
--- llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
+++ llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
@@ -75,6 +75,7 @@
 
   bool shouldExpandReduction(const IntrinsicInst *II) const;
   bool supportsScalableVectors() const { return ST->hasVInstructions(); }
+  bool enableOrderedReductions() const { return true; }
   bool enableScalableVectorization() const { return ST->hasVInstructions(); }
   TailFoldingStyle getPreferredTailFoldingStyle() const {
     return ST->hasVInstructions() ? TailFoldingStyle::Data


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D144455.499076.patch
Type: text/x-patch
Size: 4696 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230221/1da383e6/attachment.bin>