[llvm] 4bc88a0 - Enable support for floating-point division reductions

[Valentin Churavy via llvm-commits]

Author: Yichao Yu
Date: 2020-11-23T20:00:58-05:00
New Revision: 4bc88a0e9a2ee29959a9053e867ae6f051348554

URL: https://github.com/llvm/llvm-project/commit/4bc88a0e9a2ee29959a9053e867ae6f051348554
DIFF: https://github.com/llvm/llvm-project/commit/4bc88a0e9a2ee29959a9053e867ae6f051348554.diff

LOG: Enable support for floating-point division reductions

Similar to fsub, fdiv can also be vectorized using fmul.

Also http://llvm.org/viewvc/llvm-project?view=revision&revision=215200

Differential Revision: https://reviews.llvm.org/D34078

Co-authored-by: Jameson Nash <jameson at juliacomputing.com>

Added: 
    

Modified: 
    llvm/lib/Analysis/IVDescriptors.cpp
    llvm/test/Transforms/LoopVectorize/float-reduction.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Analysis/IVDescriptors.cpp b/llvm/lib/Analysis/IVDescriptors.cpp
index 86cbcb015f0f..a5cc622279c4 100644
--- a/llvm/lib/Analysis/IVDescriptors.cpp
+++ b/llvm/lib/Analysis/IVDescriptors.cpp
@@ -578,6 +578,7 @@ RecurrenceDescriptor::isRecurrenceInstr(Instruction *I, RecurrenceKind Kind,
     return InstDesc(Kind == RK_IntegerOr, I);
   case Instruction::Xor:
     return InstDesc(Kind == RK_IntegerXor, I);
+  case Instruction::FDiv:
   case Instruction::FMul:
     return InstDesc(Kind == RK_FloatMult, I, UAI);
   case Instruction::FSub:

diff  --git a/llvm/test/Transforms/LoopVectorize/float-reduction.ll b/llvm/test/Transforms/LoopVectorize/float-reduction.ll
index f3b95d0ead7d..dec854845800 100644
--- a/llvm/test/Transforms/LoopVectorize/float-reduction.ll
+++ b/llvm/test/Transforms/LoopVectorize/float-reduction.ll
@@ -44,3 +44,47 @@ for.body:                                         ; preds = %for.body, %entry
 for.end:                                          ; preds = %for.body
   ret float %sub
 }
+
+;CHECK-LABEL: @foodiv(
+;CHECK: fdiv fast <4 x float>
+;CHECK: ret
+define float @foodiv(float* nocapture %A, i32* nocapture %n) nounwind uwtable readonly ssp {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %sum.04 = phi float [ 1.000000e+00, %entry ], [ %sub, %for.body ]
+  %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
+  %0 = load float, float* %arrayidx, align 4
+  %sub = fdiv fast float %sum.04, %0
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 200
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret float %sub
+}
+
+;CHECK-LABEL: @foonodiv(
+;CHECK-NOT: fdiv fast <4 x float>
+;CHECK: ret
+define float @foonodiv(float* nocapture %A, i32* nocapture %n) nounwind uwtable readonly ssp {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %sum.04 = phi float [ 1.000000e+00, %entry ], [ %sub, %for.body ]
+  %arrayidx = getelementptr inbounds float, float* %A, i64 %indvars.iv
+  %0 = load float, float* %arrayidx, align 4
+  %sub = fdiv fast float %0, %sum.04
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 200
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret float %sub
+}