[llvm] r359313 - [PowerPC] Update P9 vector costs for insert/extract element
David Jones via llvm-commits
llvm-commits at lists.llvm.org
Tue Apr 30 21:57:00 PDT 2019
I am seeing crashes that bisect cleanly to this revision.
This is the crash reported by llc:
$ ./llc testcase-reduced.ll -o /dev/null
Stack dump:
0. Program arguments: ./llc testcase-reduced.ll -o /dev/null
1. Running pass 'Function Pass Manager' on module 'testcase-reduced.ll'.
2. Running pass 'PowerPC DAG->DAG Pattern Instruction Selection' on
function '@f'
0 llc 0x00005599bdc0bb38
1 llc 0x00005599bdc0e3e6
2 libpthread.so.0 0x00007f6114bb49a0
3 llc 0x00005599bd43e595
This looks to me like it may simply be a latent bug exposed by this
revision. It does bisect cleanly to this revision, though, so I'm going to
go ahead and revert as a temporary, stopgap measure.
Below is a fairly reduced test case (use the same command as above ^^^). It
crashes under llc as new as r359636, but works if I revert just this
revision.
I hope that the test base below is enough to make progress. Sorry about the
revert. :-(
--dlj
$ cat testcase-reduced.ll
source_filename = "375b79.cpp"
target datalayout = "e-m:e-i64:64-n32:64"
target triple = "powerpc64le-unknown-linux-gnu"
define void @f() local_unnamed_addr #0 align 32 {
br i1 undef, label %1, label %23
1: ; preds = %0
br i1 undef, label %24, label %2
2: ; preds = %1
br label %3
3: ; preds = %3, %2
%4 = phi <2 x i64> [ zeroinitializer, %2 ], [ %17, %3 ]
%5 = phi <2 x i64> [ zeroinitializer, %2 ], [ %18, %3 ]
%6 = load <2 x i32>, <2 x i32>* undef, align 4
%7 = ashr <2 x i32> %6, <i32 31, i32 31>
%8 = xor <2 x i32> zeroinitializer, %7
%9 = or <2 x i32> %8, <i32 1, i32 1>
%10 = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %9, i1 true)
%11 = xor <2 x i32> %10, <i32 31, i32 31>
%12 = mul nuw nsw <2 x i32> %11, <i32 9, i32 9>
%13 = add nuw nsw <2 x i32> %12, <i32 73, i32 73>
%14 = lshr <2 x i32> %13, <i32 6, i32 6>
%15 = zext <2 x i32> undef to <2 x i64>
%16 = zext <2 x i32> %14 to <2 x i64>
%17 = add <2 x i64> %4, %15
%18 = add <2 x i64> %5, %16
%19 = add i64 0, 24
br i1 false, label %20, label %3, !llvm.loop !1
20: ; preds = %3
%21 = add <2 x i64> %18, %17
%22 = add <2 x i64> undef, %21
br i1 undef, label %23, label %24
23: ; preds = %24, %20, %0
ret void
24: ; preds = %24, %20, %1
br i1 undef, label %23, label %24, !llvm.loop !3
}
; Function Attrs: nounwind readnone speculatable
declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1 immarg) #1
attributes #0 = { "use-soft-float"="false" }
attributes #1 = { nounwind readnone speculatable }
!llvm.ident = !{!0}
!0 = !{!"clang version eleventeen (trunk)"}
!1 = distinct !{!1, !2}
!2 = !{!"llvm.loop.isvectorized", i32 1}
!3 = distinct !{!3, !4, !2}
!4 = !{!"llvm.loop.unroll.runtime.disable"}
On Fri, Apr 26, 2019 at 9:12 AM Roland Froese via llvm-commits <
llvm-commits at lists.llvm.org> wrote:
> Author: froese
> Date: Fri Apr 26 09:14:17 2019
> New Revision: 359313
>
> URL: http://llvm.org/viewvc/llvm-project?rev=359313&view=rev
> Log:
> [PowerPC] Update P9 vector costs for insert/extract element
>
> The PPC vector cost model values for insert/extract element reflect older
> processors that lacked vector insert/extract and move-to/move-from VSR
> instructions. Update getVectorInstrCost to give appropriate values for
> when
> the newer instructions are present.
>
> Differential Revision: https://reviews.llvm.org/D60160
>
> Modified:
> llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
> llvm/trunk/test/Analysis/CostModel/PowerPC/insert_extract.ll
>
> Modified: llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp?rev=359313&r1=359312&r2=359313&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp (original)
> +++ llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp Fri Apr 26
> 09:14:17 2019
> @@ -412,6 +412,35 @@ int PPCTTIImpl::getVectorInstrCost(unsig
> return 0;
>
> return Cost;
> +
> + } else if (Val->getScalarType()->isIntegerTy() && Index != -1U) {
> + if (ST->hasP9Altivec()) {
> + if (ISD == ISD::INSERT_VECTOR_ELT)
> + // A move-to VSR and a permute/insert. Assume vector operation
> cost
> + // for both (cost will be 2x on P9).
> + return vectorCostAdjustment(2, Opcode, Val, nullptr);
> +
> + // It's an extract. Maybe we can do a cheap move-from VSR.
> + unsigned EltSize = Val->getScalarSizeInBits();
> + if (EltSize == 64) {
> + unsigned MfvsrdIndex = ST->isLittleEndian() ? 1 : 0;
> + if (Index == MfvsrdIndex)
> + return 1;
> + } else if (EltSize == 32) {
> + unsigned MfvsrwzIndex = ST->isLittleEndian() ? 2 : 1;
> + if (Index == MfvsrwzIndex)
> + return 1;
> + }
> +
> + // We need a vector extract (or mfvsrld). Assume vector operation
> cost.
> + // The cost of the load constant for a vector extract is disregarded
> + // (invariant, easily schedulable).
> + return vectorCostAdjustment(1, Opcode, Val, nullptr);
> +
> + } else if (ST->hasDirectMove())
> + // Assume permute has standard cost.
> + // Assume move-to/move-from VSR have 2x standard cost.
> + return 3;
> }
>
> // Estimated cost of a load-hit-store delay. This was obtained
>
> Modified: llvm/trunk/test/Analysis/CostModel/PowerPC/insert_extract.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/PowerPC/insert_extract.ll?rev=359313&r1=359312&r2=359313&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/Analysis/CostModel/PowerPC/insert_extract.ll (original)
> +++ llvm/trunk/test/Analysis/CostModel/PowerPC/insert_extract.ll Fri Apr
> 26 09:14:17 2019
> @@ -14,15 +14,15 @@ define i32 @insert(i32 %arg) {
> ; CHECK-P7-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret i32 undef
> ;
> ; CHECK-P8LE-LABEL: 'insert'
> -; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 10 for
> instruction: %x = insertelement <4 x i32> undef, i32 %arg, i32 0
> +; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 3 for
> instruction: %x = insertelement <4 x i32> undef, i32 %arg, i32 0
> ; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret i32 undef
> ;
> ; CHECK-P9BE-LABEL: 'insert'
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 11 for
> instruction: %x = insertelement <4 x i32> undef, i32 %arg, i32 0
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %x = insertelement <4 x i32> undef, i32 %arg, i32 0
> ; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret i32 undef
> ;
> ; CHECK-P9LE-LABEL: 'insert'
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 11 for
> instruction: %x = insertelement <4 x i32> undef, i32 %arg, i32 0
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %x = insertelement <4 x i32> undef, i32 %arg, i32 0
> ; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret i32 undef
> ;
> %x = insertelement <4 x i32> undef, i32 %arg, i32 0
> @@ -40,11 +40,11 @@ define i32 @extract(<4 x i32> %arg) {
> ; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret i32 %x
> ;
> ; CHECK-P9BE-LABEL: 'extract'
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %x = extractelement <4 x i32> %arg, i32 0
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %x = extractelement <4 x i32> %arg, i32 0
> ; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret i32 %x
> ;
> ; CHECK-P9LE-LABEL: 'extract'
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %x = extractelement <4 x i32> %arg, i32 0
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %x = extractelement <4 x i32> %arg, i32 0
> ; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret i32 %x
> ;
> %x = extractelement <4 x i32> %arg, i32 0
> @@ -83,15 +83,15 @@ define void @test4xi32(<4 x i32> %v1, i3
> ; CHECK-P7-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P8LE-LABEL: 'test4xi32'
> -; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 10 for
> instruction: %v2 = insertelement <4 x i32> %v1, i32 %x1, i32 2
> +; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 3 for
> instruction: %v2 = insertelement <4 x i32> %v1, i32 %x1, i32 2
> ; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P9BE-LABEL: 'test4xi32'
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 11 for
> instruction: %v2 = insertelement <4 x i32> %v1, i32 %x1, i32 2
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %v2 = insertelement <4 x i32> %v1, i32 %x1, i32 2
> ; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P9LE-LABEL: 'test4xi32'
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 11 for
> instruction: %v2 = insertelement <4 x i32> %v1, i32 %x1, i32 2
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %v2 = insertelement <4 x i32> %v1, i32 %x1, i32 2
> ; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> %v2 = insertelement <4 x i32> %v1, i32 %x1, i32 2
> @@ -114,17 +114,17 @@ define void @vexti32(<4 x i32> %p1) {
> ; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P9BE-LABEL: 'vexti32'
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i1 = extractelement <4 x i32> %p1, i32 0
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i2 = extractelement <4 x i32> %p1, i32 1
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i3 = extractelement <4 x i32> %p1, i32 2
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i4 = extractelement <4 x i32> %p1, i32 3
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i1 = extractelement <4 x i32> %p1, i32 0
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 1 for
> instruction: %i2 = extractelement <4 x i32> %p1, i32 1
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i3 = extractelement <4 x i32> %p1, i32 2
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i4 = extractelement <4 x i32> %p1, i32 3
> ; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P9LE-LABEL: 'vexti32'
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i1 = extractelement <4 x i32> %p1, i32 0
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i2 = extractelement <4 x i32> %p1, i32 1
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i3 = extractelement <4 x i32> %p1, i32 2
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i4 = extractelement <4 x i32> %p1, i32 3
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i1 = extractelement <4 x i32> %p1, i32 0
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i2 = extractelement <4 x i32> %p1, i32 1
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 1 for
> instruction: %i3 = extractelement <4 x i32> %p1, i32 2
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i4 = extractelement <4 x i32> %p1, i32 3
> ; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> %i1 = extractelement <4 x i32> %p1, i32 0
> @@ -146,13 +146,13 @@ define void @vexti64(<2 x i64> %p1) {
> ; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P9BE-LABEL: 'vexti64'
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i1 = extractelement <2 x i64> %p1, i32 0
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i2 = extractelement <2 x i64> %p1, i32 1
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 1 for
> instruction: %i1 = extractelement <2 x i64> %p1, i32 0
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i2 = extractelement <2 x i64> %p1, i32 1
> ; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P9LE-LABEL: 'vexti64'
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i1 = extractelement <2 x i64> %p1, i32 0
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i2 = extractelement <2 x i64> %p1, i32 1
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i1 = extractelement <2 x i64> %p1, i32 0
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 1 for
> instruction: %i2 = extractelement <2 x i64> %p1, i32 1
> ; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> %i1 = extractelement <2 x i64> %p1, i32 0
> @@ -172,13 +172,13 @@ define void @vext(<8 x i16> %p1, <16 x i
> ; CHECK-P8LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P9BE-LABEL: 'vext'
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i1 = extractelement <8 x i16> %p1, i32 0
> -; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i2 = extractelement <16 x i8> %p2, i32 0
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i1 = extractelement <8 x i16> %p1, i32 0
> +; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i2 = extractelement <16 x i8> %p2, i32 0
> ; CHECK-P9BE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> ; CHECK-P9LE-LABEL: 'vext'
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i1 = extractelement <8 x i16> %p1, i32 0
> -; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 4 for
> instruction: %i2 = extractelement <16 x i8> %p2, i32 0
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i1 = extractelement <8 x i16> %p1, i32 0
> +; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 2 for
> instruction: %i2 = extractelement <16 x i8> %p2, i32 0
> ; CHECK-P9LE-NEXT: Cost Model: Found an estimated cost of 0 for
> instruction: ret void
> ;
> %i1 = extractelement <8 x i16> %p1, i32 0
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190430/ca8b2969/attachment.html>
More information about the llvm-commits
mailing list