[llvm] r340947 - [InstCombine] Expand the simplification of pow() into exp2()
Reid Kleckner via llvm-commits
llvm-commits at lists.llvm.org
Thu Aug 30 11:56:50 PDT 2018
I don't have time to try it, but I made this reduction:
target datalayout = "e-m:w-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-windows-msvc19.11.0"
@a = dso_local global double 0.000000e+00, align 8
define dso_local double @b() {
entry:
%0 = load double, double* @a, align 8
%call = call double @pow(double 2.000000e+00, double %0)
ret double %call
}
declare dso_local double @pow(double, double)
I suspect the triple is the important part.
On Thu, Aug 30, 2018 at 11:01 AM Evandro Menezes <e.menezes at samsung.com>
wrote:
> Hey, Reid.
>
> I don't have access to a Windows machine. Would it be possible for you
> to try the new patch at https://reviews.llvm.org/D49273?
>
> Thank you,
>
> --
> Evandro Menezes
>
>
> On 08/29/2018 06:19 PM, Reid Kleckner wrote:
>
> It's a name LLVM makes up when mangling unnamed private globals like @0 =
> global i32. I'm not yet sure how your change causes these problems, but
> I've gone ahead and reverted your change in r340991 and am starting a
> reduction now. I have to run soon so I won't be able to share it until
> tomorrow, most likely.
>
> On Wed, Aug 29, 2018 at 4:05 PM Evandro Menezes <e.menezes at samsung.com>
> wrote:
>
>> Reid,
>>
>> Can you please provide an indication of what this symbol is about, as I'm
>> not familiar with the VS linker?
>>
>> Thank you,
>>
>> --
>> Evandro Menezes
>>
>>
>> On 08/29/2018 05:50 PM, Reid Kleckner wrote:
>>
>> My bisection shows that this broke the clang stage 2 build:
>> http://lab.llvm.org:8011/builders/clang-x86-windows-msvc2015/builds/13525
>>
>> We end up with references to __unnamed_1 globals that are unresolved.
>> I'll revert and get a reproducer.
>>
>> On Wed, Aug 29, 2018 at 11:00 AM Evandro Menezes via llvm-commits <
>> llvm-commits at lists.llvm.org> wrote:
>>
>>> Author: evandro
>>> Date: Wed Aug 29 10:59:34 2018
>>> New Revision: 340947
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=340947&view=rev
>>> Log:
>>> [InstCombine] Expand the simplification of pow() into exp2()
>>>
>>> Generalize the simplification of `pow(2.0, y)` to `pow(2.0 ** n, y)` for
>>> all
>>> scalar and vector types.
>>>
>>> This improvement helps some benchmarks in SPEC CPU2000 and CPU2006, such
>>> as
>>> 252.eon, 447.dealII, 453.povray. Otherwise, no significant regressions
>>> on
>>> x86-64 or A64.
>>>
>>> Differential revision: https://reviews.llvm.org/D49273
>>>
>>> Modified:
>>> llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp
>>> llvm/trunk/test/Transforms/InstCombine/pow-1.ll
>>>
>>> Modified: llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp
>>> URL:
>>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp?rev=340947&r1=340946&r2=340947&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp (original)
>>> +++ llvm/trunk/lib/Transforms/Utils/SimplifyLibCalls.cpp Wed Aug 29
>>> 10:59:34 2018
>>> @@ -13,6 +13,7 @@
>>>
>>> //===----------------------------------------------------------------------===//
>>>
>>> #include "llvm/Transforms/Utils/SimplifyLibCalls.h"
>>> +#include "llvm/ADT/APSInt.h"
>>> #include "llvm/ADT/SmallString.h"
>>> #include "llvm/ADT/StringMap.h"
>>> #include "llvm/ADT/Triple.h"
>>> @@ -1183,12 +1184,13 @@ static Value *getPow(Value *InnerChain[3
>>> }
>>>
>>> /// Use exp{,2}(x * y) for pow(exp{,2}(x), y);
>>> -/// exp2(x) for pow(2.0, x); exp10(x) for pow(10.0, x).
>>> +/// exp2(n * x) for pow(2.0 ** n, x); exp10(x) for pow(10.0, x).
>>> Value *LibCallSimplifier::replacePowWithExp(CallInst *Pow, IRBuilder<>
>>> &B) {
>>> Value *Base = Pow->getArgOperand(0), *Expo = Pow->getArgOperand(1);
>>> AttributeList Attrs = Pow->getCalledFunction()->getAttributes();
>>> Module *Mod = Pow->getModule();
>>> Type *Ty = Pow->getType();
>>> + bool Ignored;
>>>
>>> // Evaluate special cases related to a nested function as the base.
>>>
>>> @@ -1232,10 +1234,28 @@ Value *LibCallSimplifier::replacePowWith
>>>
>>> // Evaluate special cases related to a constant base.
>>>
>>> - // pow(2.0, x) -> exp2(x)
>>> - if (match(Base, m_SpecificFP(2.0))) {
>>> - Value *Exp2 = Intrinsic::getDeclaration(Mod, Intrinsic::exp2, Ty);
>>> - return B.CreateCall(Exp2, Expo, "exp2");
>>> + const APFloat *BaseF;
>>> + if (!match(Pow->getArgOperand(0), m_APFloat(BaseF)))
>>> + return nullptr;
>>> +
>>> + // pow(2.0 ** n, x) -> exp2(n * x)
>>> + APFloat BaseR = APFloat(1.0);
>>> + BaseR.convert(BaseF->getSemantics(), APFloat::rmTowardZero, &Ignored);
>>> + BaseR = BaseR / *BaseF;
>>> + bool IsInteger = BaseF->isInteger(),
>>> + IsReciprocal = BaseR.isInteger();
>>> + const APFloat *NF = IsReciprocal ? &BaseR : BaseF;
>>> + APSInt NI(64, false);
>>> + if ((IsInteger || IsReciprocal) &&
>>> + !NF->convertToInteger(NI, APFloat::rmTowardZero, &Ignored) &&
>>> + NI > 1 && NI.isPowerOf2()) {
>>> + double N = NI.logBase2() * (IsReciprocal ? -1.0 : 1.0);
>>> + Value *FMul = B.CreateFMul(Expo, ConstantFP::get(Ty, N), "mul");
>>> + if (Pow->doesNotAccessMemory())
>>> + return B.CreateCall(Intrinsic::getDeclaration(Mod,
>>> Intrinsic::exp2, Ty),
>>> + FMul, "exp2");
>>> + else
>>> + return emitUnaryFloatFnCall(FMul, TLI->getName(LibFunc_exp2), B,
>>> Attrs);
>>> }
>>>
>>> // pow(10.0, x) -> exp10(x)
>>>
>>> Modified: llvm/trunk/test/Transforms/InstCombine/pow-1.ll
>>> URL:
>>> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/pow-1.ll?rev=340947&r1=340946&r2=340947&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/Transforms/InstCombine/pow-1.ll (original)
>>> +++ llvm/trunk/test/Transforms/InstCombine/pow-1.ll Wed Aug 29 10:59:34
>>> 2018
>>> @@ -16,6 +16,7 @@
>>>
>>> declare float @powf(float, float) nounwind readonly
>>> declare double @pow(double, double) nounwind readonly
>>> +declare double @llvm.pow.f64(double, double)
>>> declare <2 x float> @llvm.pow.v2f32(<2 x float>, <2 x float>) nounwind
>>> readonly
>>> declare <2 x double> @llvm.pow.v2f64(<2 x double>, <2 x double>)
>>> nounwind readonly
>>>
>>> @@ -57,18 +58,18 @@ define <2 x double> @test_simplify2v(<2
>>>
>>> define float @test_simplify3(float %x) {
>>> ; ANY-LABEL: @test_simplify3(
>>> -; ANY-NEXT: [[EXP2:%.*]] = call float @llvm.exp2.f32(float [[X:%.*]])
>>> -; ANY-NEXT: ret float [[EXP2]]
>>> +; ANY-NEXT: [[EXP2F:%.*]] = call float @exp2f(float [[X:%.*]])
>>> [[NUW_RO:#[0-9]+]]
>>> +; ANY-NEXT: ret float [[EXP2F]]
>>> ;
>>> %retval = call float @powf(float 2.0, float %x)
>>> ret float %retval
>>> }
>>>
>>> -; TODO: Should result in exp2(-2.0 * x).
>>> define double @test_simplify3n(double %x) {
>>> ; ANY-LABEL: @test_simplify3n(
>>> -; ANY-NEXT: [[RETVAL:%.*]] = call double @pow(double 2.500000e-01,
>>> double [[X:%.*]])
>>> -; ANY-NEXT: ret double [[RETVAL]]
>>> +; ANY-NEXT: [[MUL:%.*]] = fmul double [[X:%.*]], -2.000000e+00
>>> +; ANY-NEXT: [[EXP2:%.*]] = call double @exp2(double [[MUL]])
>>> [[NUW_RO]]
>>> +; ANY-NEXT: ret double [[EXP2]]
>>> ;
>>> %retval = call double @pow(double 0.25, double %x)
>>> ret double %retval
>>> @@ -83,11 +84,11 @@ define <2 x float> @test_simplify3v(<2 x
>>> ret <2 x float> %retval
>>> }
>>>
>>> -; TODO: Should result in exp2(2.0 * x).
>>> define <2 x double> @test_simplify3vn(<2 x double> %x) {
>>> ; ANY-LABEL: @test_simplify3vn(
>>> -; ANY-NEXT: [[RETVAL:%.*]] = call <2 x double> @llvm.pow.v2f64(<2 x
>>> double> <double 4.000000e+00, double 4.000000e+00>, <2 x double> [[X:%.*]])
>>> -; ANY-NEXT: ret <2 x double> [[RETVAL]]
>>> +; ANY-NEXT: [[MUL:%.*]] = fmul <2 x double> [[X:%.*]], <double
>>> 2.000000e+00, double 2.000000e+00>
>>> +; ANY-NEXT: [[EXP2:%.*]] = call <2 x double> @llvm.exp2.v2f64(<2 x
>>> double> [[MUL]])
>>> +; ANY-NEXT: ret <2 x double> [[EXP2]]
>>> ;
>>> %retval = call <2 x double> @llvm.pow.v2f64(<2 x double> <double 4.0,
>>> double 4.0>, <2 x double> %x)
>>> ret <2 x double> %retval
>>> @@ -95,18 +96,18 @@ define <2 x double> @test_simplify3vn(<2
>>>
>>> define double @test_simplify4(double %x) {
>>> ; ANY-LABEL: @test_simplify4(
>>> -; ANY-NEXT: [[EXP2:%.*]] = call double @llvm.exp2.f64(double
>>> [[X:%.*]])
>>> +; ANY-NEXT: [[EXP2:%.*]] = call double @exp2(double [[X:%.*]])
>>> [[NUW_RO]]
>>> ; ANY-NEXT: ret double [[EXP2]]
>>> ;
>>> %retval = call double @pow(double 2.0, double %x)
>>> ret double %retval
>>> }
>>>
>>> -; TODO: Should result in exp2f(3.0 * x).
>>> define float @test_simplify4n(float %x) {
>>> ; ANY-LABEL: @test_simplify4n(
>>> -; ANY-NEXT: [[RETVAL:%.*]] = call float @powf(float 8.000000e+00,
>>> float [[X:%.*]])
>>> -; ANY-NEXT: ret float [[RETVAL]]
>>> +; ANY-NEXT: [[MUL:%.*]] = fmul float [[X:%.*]], 3.000000e+00
>>> +; ANY-NEXT: [[EXP2F:%.*]] = call float @exp2f(float [[MUL]])
>>> [[NUW_RO]]
>>> +; ANY-NEXT: ret float [[EXP2F]]
>>> ;
>>> %retval = call float @powf(float 8.0, float %x)
>>> ret float %retval
>>> @@ -121,11 +122,11 @@ define <2 x double> @test_simplify4v(<2
>>> ret <2 x double> %retval
>>> }
>>>
>>> -; TODO: Should result in exp2f(-x).
>>> define <2 x float> @test_simplify4vn(<2 x float> %x) {
>>> ; ANY-LABEL: @test_simplify4vn(
>>> -; ANY-NEXT: [[RETVAL:%.*]] = call <2 x float> @llvm.pow.v2f32(<2 x
>>> float> <float 5.000000e-01, float 5.000000e-01>, <2 x float> [[X:%.*]])
>>> -; ANY-NEXT: ret <2 x float> [[RETVAL]]
>>> +; ANY-NEXT: [[MUL:%.*]] = fsub <2 x float> <float -0.000000e+00,
>>> float -0.000000e+00>, [[X:%.*]]
>>> +; ANY-NEXT: [[EXP2:%.*]] = call <2 x float> @llvm.exp2.v2f32(<2 x
>>> float> [[MUL]])
>>> +; ANY-NEXT: ret <2 x float> [[EXP2]]
>>> ;
>>> %retval = call <2 x float> @llvm.pow.v2f32(<2 x float> <float 0.5,
>>> float 0.5>, <2 x float> %x)
>>> ret <2 x float> %retval
>>> @@ -169,7 +170,7 @@ define <2 x double> @test_simplify6v(<2
>>>
>>> define float @test_simplify7(float %x) {
>>> ; ANY-LABEL: @test_simplify7(
>>> -; ANY-NEXT: [[SQRTF:%.*]] = call float @sqrtf(float [[X:%.*]])
>>> [[NUW_RO:#[0-9]+]]
>>> +; ANY-NEXT: [[SQRTF:%.*]] = call float @sqrtf(float [[X:%.*]])
>>> [[NUW_RO]]
>>> ; ANY-NEXT: [[ABS:%.*]] = call float @llvm.fabs.f32(float [[SQRTF]])
>>> ; ANY-NEXT: [[ISINF:%.*]] = fcmp oeq float [[X]], 0xFFF0000000000000
>>> ; ANY-NEXT: [[TMP1:%.*]] = select i1 [[ISINF]], float
>>> 0x7FF0000000000000, float [[ABS]]
>>> @@ -330,7 +331,6 @@ define <2 x double> @pow_neg1_double_fas
>>> ret <2 x double> %r
>>> }
>>>
>>> -declare double @llvm.pow.f64(double %Val, double %Power)
>>> define double @test_simplify17(double %x) {
>>> ; ANY-LABEL: @test_simplify17(
>>> ; ANY-NEXT: [[SQRT:%.*]] = call double @llvm.sqrt.f64(double
>>> [[X:%.*]])
>>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at lists.llvm.org
>>> http://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/20180830/b55ca6df/attachment.html>
More information about the llvm-commits
mailing list