[llvm] r185299 - LoopVectorize: Math functions only read rounding mode

Duncan Sands duncan.sands at gmail.com
Mon Jul 1 07:35:15 PDT 2013


Hi Arnold,

On 01/07/13 02:54, Arnold Schwaighofer wrote:
> Author: arnolds
> Date: Sun Jun 30 19:54:44 2013
> New Revision: 185299
>
> URL: http://llvm.org/viewvc/llvm-project?rev=185299&view=rev
> Log:
> LoopVectorize: Math functions only read rounding mode
>
> Math functions are mark as readonly because they read the floating point
> rounding mode. Because we don't vectorize loops that would contain function
> calls that set the rounding mode it is safe to ignore this memory read.

how do you know that the intrinsic is a math intrinsic?  For example it could
be an x86 unaligned load intrinsic, vector load intrinsic, or who knows what,
that may be reading from some memory that is written to in the loop.

Ciao, Duncan.

>
> Added:
>      llvm/trunk/test/Transforms/LoopVectorize/funcall.ll
> Modified:
>      llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp
>
> Modified: llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp?rev=185299&r1=185298&r2=185299&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp (original)
> +++ llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp Sun Jun 30 19:54:44 2013
> @@ -3520,6 +3520,13 @@ bool LoopVectorizationLegality::canVecto
>         // but is not a load, then we quit. Notice that we don't handle function
>         // calls that read or write.
>         if (it->mayReadFromMemory()) {
> +        // Many math library functions read the rounding mode. We will only
> +        // vectorize a loop if it contains known function calls that don't set
> +        // the flag. Therefore, it is safe to ignore this read from memory.
> +        CallInst *Call = dyn_cast<CallInst>(it);
> +        if (Call && getIntrinsicIDForCall(Call, TLI))
> +          continue;
> +
>           LoadInst *Ld = dyn_cast<LoadInst>(it);
>           if (!Ld) return false;
>           if (!Ld->isSimple() && !IsAnnotatedParallel) {
>
> Added: llvm/trunk/test/Transforms/LoopVectorize/funcall.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/funcall.ll?rev=185299&view=auto
> ==============================================================================
> --- llvm/trunk/test/Transforms/LoopVectorize/funcall.ll (added)
> +++ llvm/trunk/test/Transforms/LoopVectorize/funcall.ll Sun Jun 30 19:54:44 2013
> @@ -0,0 +1,32 @@
> +; RUN: opt -S -loop-vectorize -force-vector-width=2 -force-vector-unroll=1 < %s | FileCheck %s
> +
> +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
> +
> +; Make sure we can vectorize loops with functions to math library functions.
> +; They might read the rounding mode but we are only vectorizing loops that
> +; contain a limited set of function calls and none of them sets the rounding
> +; mode, so vectorizing them is safe.
> +
> +; CHECK: test
> +; CHECK: <2 x double>
> +
> +define void @test(double* %d, double %t) {
> +entry:
> +  br label %for.body
> +
> +for.body:
> +  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
> +  %arrayidx = getelementptr inbounds double* %d, i64 %indvars.iv
> +  %0 = load double* %arrayidx, align 8
> +  %1 = tail call double @llvm.pow.f64(double %0, double %t)
> +  store double %1, double* %arrayidx, align 8
> +  %indvars.iv.next = add i64 %indvars.iv, 1
> +  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
> +  %exitcond = icmp ne i32 %lftr.wideiv, 128
> +  br i1 %exitcond, label %for.body, label %for.end
> +
> +for.end:
> +  ret void
> +}
> +
> +declare double @llvm.pow.f64(double, double)
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>




More information about the llvm-commits mailing list