[SLC] Expand the simplification of pow(x, 0.5) to sqrt(x)

[evandro]

Summary

Expand the number of cases when pow(x, 0.5) is simplified into sqrt(x) by considering the math semantics with more granularity.

[evandro]

Ping! 🔔

[spatel]

llvm/test/Transforms/InstCombine/pow-sqrt.ll

8–10

Why change the data types?

20

The minimal test for this case requires no FMF?

30–31

Why change the data types? Same question for the later tests - the types should be irrelevant to the change in functionality, right?

Please make any IR changes to the tests that are necessary to show off the improved/fixed functionality as a preliminary step to this review.

[evandro]

I'll commit the data type changes increasing type diversity on the side.

llvm/test/Transforms/InstCombine/pow-sqrt.ll

8–10

In order to increase test coverage. Otherwise, it virtually tested for double only.

20

IIUC, not all of them, just AFN.

30–31

Replacing AFN for NINF shows the improved functionality.

[lebedev.ri]

llvm/test/Transforms/InstCombine/pow-sqrt.ll

8–10

Tests are cheap. Couldn't you just *add* the additional test coverage with floats?
Changing existing tests is not always the best way..

[evandro]

llvm/test/Transforms/InstCombine/pow-sqrt.ll

8–10

Yes, you're right.

[spatel]

llvm/test/Transforms/InstCombine/pow-sqrt.ll

20

'afn' means this transform can diverge from the original code in some way, but I don't think that's true here. Ie, sqrt(x) should be the same as pow(x, 0.5) for all values other than -0.0 and -INFINITY.
Not sure if that is specified by any standard though? (cc @scanon)

[evandro]

llvm/test/Transforms/InstCombine/pow-sqrt.ll

20

The test originally had AFN, so I aimed at preserving it.

[spatel]

llvm/test/Transforms/InstCombine/pow-sqrt.ll

20

It's not obvious where the holes are in the existing code, so I added tests at:
rL339711
rL339713 (fix copy/paste errors)

If you want to vary the data types more in those tests, feel free. But I think that should provide coverage for everything that is changing in this patch based on FMF + libcall vs. intrinsic. Please rebase this patch with those tests.

[spatel]

LGTM

[evandro]

Thank you.

llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp

static Value *getPow(Value *InnerChain[33], unsigned Exp, IRBuilder<> &B) {
 
   InnerChain[Exp] = B.CreateFMul(getPow(InnerChain, AddChain[Exp][0], B),
                                  getPow(InnerChain, AddChain[Exp][1], B));
   return InnerChain[Exp];
 }
 
 /// Use square root in place of pow(x, +/-0.5).
 Value *LibCallSimplifier::replacePowWithSqrt(CallInst *Pow, IRBuilder<> &B) {
-  // TODO: There is some subset of 'fast' under which these transforms should
-  // be allowed.
-  if (!Pow->isFast())
-    return nullptr;
-
   Value *Sqrt, *Base = Pow->getArgOperand(0), *Expo = Pow->getArgOperand(1);
+  Module *Mod = Pow->getModule();
   Type *Ty = Pow->getType();
 
   const APFloat *ExpoF;
   if (!match(Expo, m_APFloat(ExpoF)) ||
       (!ExpoF->isExactlyValue(0.5) && !ExpoF->isExactlyValue(-0.5)))
     return nullptr;
 
   // If errno is never set, then use the intrinsic for sqrt().
   if (Pow->hasFnAttr(Attribute::ReadNone)) {
     Function *SqrtFn = Intrinsic::getDeclaration(Pow->getModule(),
                                                  Intrinsic::sqrt, Ty);
-    Sqrt = B.CreateCall(SqrtFn, Base);
+    Sqrt = B.CreateCall(SqrtFn, Base, "sqrt");
   }
   // Otherwise, use the libcall for sqrt().
   else if (hasUnaryFloatFn(TLI, Ty, LibFunc_sqrt, LibFunc_sqrtf, LibFunc_sqrtl))
     // TODO: We also should check that the target can in fact lower the sqrt()
     // libcall. We currently have no way to ask this question, so we ask if
     // the target has a sqrt() libcall, which is not exactly the same.
     Sqrt = emitUnaryFloatFnCall(Base, TLI->getName(LibFunc_sqrt), B,
                                 Pow->getCalledFunction()->getAttributes());
   else
     return nullptr;
 
+  // Handle signed zero base by expanding to fabs(sqrt(x)).
+  if (!Pow->hasNoSignedZeros()) {
+    Function *FAbsFn = Intrinsic::getDeclaration(Mod, Intrinsic::fabs, Ty);
+    Sqrt = B.CreateCall(FAbsFn, Sqrt, "abs");
+  }
+
+  // Handle non finite base by expanding to
+  // (x == -infinity ? +infinity : sqrt(x)).
+  if (!Pow->hasNoInfs()) {
+    Value *PosInf = ConstantFP::getInfinity(Ty),
+          *NegInf = ConstantFP::getInfinity(Ty, true);
+    Value *FCmp = B.CreateFCmpOEQ(Base, NegInf, "iseq");
+    Sqrt = B.CreateSelect(FCmp, PosInf, Sqrt);
+  }
+
   // If the exponent is negative, then get the reciprocal.
   if (ExpoF->isNegative())
     Sqrt = B.CreateFDiv(ConstantFP::get(Ty, 1.0), Sqrt, "reciprocal");
 
   return Sqrt;
 }
 
 Value *LibCallSimplifier::optimizePow(CallInst *Pow, IRBuilder<> &B) {
if (BaseC->isExactlyValue(10.0) &&
         hasUnaryFloatFn(TLI, Ty, LibFunc_exp10, LibFunc_exp10f, LibFunc_exp10l))
       return emitUnaryFloatFnCall(Expo, TLI->getName(LibFunc_exp10), B, Attrs);
 
   // pow(exp(x), y) -> exp(x * y)
   // pow(exp2(x), y) -> exp2(x * y)
   // We enable these only with fast-math. Besides rounding differences, the
   // transformation changes overflow and underflow behavior quite dramatically.
   // Example: x = 1000, y = 0.001.
-  // pow(exp(x), y) = pow(inf, 0.001) = inf, whereas exp(x*y) = exp(1).
+  // pow(exp(x), y) = pow(inf, 0.001) = inf, whereas exp(x * y) = exp(1).
   auto *BaseFn = dyn_cast<CallInst>(Base);
   if (BaseFn && BaseFn->isFast() && Pow->isFast()) {
     LibFunc LibFn;
     Function *CalleeFn = BaseFn->getCalledFunction();
     if (CalleeFn && TLI->getLibFunc(CalleeFn->getName(), LibFn) &&
         (LibFn == LibFunc_exp || LibFn == LibFunc_exp2) && TLI->has(LibFn)) {
       Value *FMul = B.CreateFMul(BaseFn->getArgOperand(0), Expo, "mul");
       return emitUnaryFloatFnCall(FMul, CalleeFn->getName(), B,
Value *LibCallSimplifier::optimizePow(CallInst *Pow, IRBuilder<> &B) {
 
   // pow(x, 2.0) -> x * x
   if (match(Expo, m_SpecificFP(2.0)))
     return B.CreateFMul(Base, Base, "square");
 
   if (Value *Sqrt = replacePowWithSqrt(Pow, B))
     return Sqrt;
 
-  // FIXME: Correct the transforms and pull this into replacePowWithSqrt().
-  ConstantFP *ExpoC = dyn_cast<ConstantFP>(Expo);
-  if (ExpoC && ExpoC->isExactlyValue(0.5) &&
-      hasUnaryFloatFn(TLI, Ty, LibFunc_sqrt, LibFunc_sqrtf, LibFunc_sqrtl)) {
-    // Expand pow(x, 0.5) to (x == -infinity ? +infinity : fabs(sqrt(x))).
-    // This is faster than calling pow(), and still handles -0.0 and
-    // negative infinity correctly.
-    // TODO: In finite-only mode, this could be just fabs(sqrt(x)).
-    Value *PosInf = ConstantFP::getInfinity(Ty);
-    Value *NegInf = ConstantFP::getInfinity(Ty, true);
-
-    // TODO: As above, we should lower to the sqrt intrinsic if the pow is an
-    // intrinsic, to match errno semantics.
-    Value *Sqrt = emitUnaryFloatFnCall(Base, TLI->getName(LibFunc_sqrt),
-                                       B, Attrs);
-    Function *FAbsFn = Intrinsic::getDeclaration(Module, Intrinsic::fabs, Ty);
-    Value *FAbs = B.CreateCall(FAbsFn, Sqrt, "abs");
-    Value *FCmp = B.CreateFCmpOEQ(Base, NegInf);
-    Sqrt = B.CreateSelect(FCmp, PosInf, FAbs);
-    return Sqrt;
-  }
-
   // pow(x, n) -> x * x * x * ...
   const APFloat *ExpoF;
   if (Pow->isFast() && match(Expo, m_APFloat(ExpoF))) {
     // We limit to a max of 7 multiplications, thus the maximum exponent is 32.
     APFloat LimF(ExpoF->getSemantics(), 33.0),
             ExpoA(abs(*ExpoF));
     if (ExpoA.isInteger() && ExpoA.compare(LimF) == APFloat::cmpLessThan) {
       // We will memoize intermediate products of the Addition Chain.

llvm/test/Transforms/InstCombine/pow-1.ll

;
   %r = call fast double @pow(double %x, double -1.0)
   ret double %r
 }
 
 declare double @llvm.pow.f64(double %Val, double %Power)
 define double @test_simplify17(double %x) {
 ; CHECK-LABEL: @test_simplify17(
   %retval = call double @llvm.pow.f64(double %x, double 0.5)
-; CHECK-NEXT: [[SQRT:%[a-z0-9]+]] = call double @sqrt(double %x)
+; CHECK-NEXT: [[SQRT:%[a-z0-9]+]] = call double @llvm.sqrt.f64(double %x)
 ; CHECK-NEXT: [[FABS:%[a-z0-9]+]] = call double @llvm.fabs.f64(double [[SQRT]])
 ; CHECK-NEXT: [[FCMP:%[a-z0-9]+]] = fcmp oeq double %x, 0xFFF0000000000000
 ; CHECK-NEXT: [[SELECT:%[a-z0-9]+]] = select i1 [[FCMP]], double 0x7FF0000000000000, double [[FABS]]
   ret double %retval
 ; CHECK-NEXT: ret double [[SELECT]]
 }
 
 ; Check pow(10.0, x) -> __exp10(x) on OS X 10.9+ and iOS 7.0+.

llvm/test/Transforms/InstCombine/pow-sqrt.ll

 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
-define double @pow_intrinsic_half_fast(double %x) {
-; CHECK-LABEL: @pow_intrinsic_half_fast(
-; CHECK-NEXT:    [[TMP1:%.*]] = call fast double @llvm.sqrt.f64(double %x)
-; CHECK-NEXT:    ret double [[TMP1]]
+define float @powf_intrinsic_half_fast(float %x) {
+; CHECK-LABEL: @powf_intrinsic_half_fast(
+; CHECK-NEXT:    [[SQRT:%.*]] = call fast float @llvm.sqrt.f32(float %x)
+; CHECK-NEXT:    ret float [[SQRT]]
 ;
-  %pow = call fast double @llvm.pow.f64(double %x, double 5.000000e-01)
-  ret double %pow
+  %pow = call fast float @llvm.pow.f32(float %x, float 5.0e-01)
+  ret float %pow
 }
 
 define <2 x double> @pow_intrinsic_half_approx(<2 x double> %x) {
 ; CHECK-LABEL: @pow_intrinsic_half_approx(
-; CHECK-NEXT:    [[POW:%.*]] = call afn <2 x double> @llvm.pow.v2f64(<2 x double> %x, <2 x double> <double 5.000000e-01, double 5.000000e-01>)
-; CHECK-NEXT:    ret <2 x double> [[POW]]
+; CHECK-NEXT:    [[SQRT:%.*]] = call afn <2 x double> @llvm.sqrt.v2f64(<2 x double> %x)
+; CHECK-NEXT:    [[TMP1:%.*]] = call afn <2 x double> @llvm.fabs.v2f64(<2 x double> [[SQRT]])
+; CHECK-NEXT:    [[TMP2:%.*]] = fcmp afn oeq <2 x double> %x, <double 0xFFF0000000000000, double 0xFFF0000000000000>
+; CHECK-NEXT:    [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x double> <double 0x7FF0000000000000, double 0x7FF0000000000000>, <2 x double> [[TMP1]]
+; CHECK-NEXT:    ret <2 x double> [[TMP3]]
 ;
   %pow = call afn <2 x double> @llvm.pow.v2f64(<2 x double> %x, <2 x double> <double 5.0e-01, double 5.0e-01>)
   ret <2 x double> %pow
 }
 
-define double @pow_libcall_half_approx(double %x) {
-; CHECK-LABEL: @pow_libcall_half_approx(
-; CHECK-NEXT:    [[SQRT:%.*]] = call afn double @sqrt(double %x)
-; CHECK-NEXT:    [[TMP1:%.*]] = call afn double @llvm.fabs.f64(double [[SQRT]])
-; CHECK-NEXT:    [[TMP2:%.*]] = fcmp afn oeq double %x, 0xFFF0000000000000
-; CHECK-NEXT:    [[TMP3:%.*]] = select i1 [[TMP2]], double 0x7FF0000000000000, double [[TMP1]]
-; CHECK-NEXT:    ret double [[TMP3]]
+define float @powf_libcall_half_ninf(float %x) {
+; CHECK-LABEL: @powf_libcall_half_ninf(
+; CHECK-NEXT:    [[SQRTF:%.*]] = call ninf float @sqrtf(float %x)
+; CHECK-NEXT:    [[TMP1:%.*]] = call ninf float @llvm.fabs.f32(float [[SQRTF]])
+; CHECK-NEXT:    ret float [[TMP1]]
 ;
-  %pow = call afn double @pow(double %x, double 5.0e-01)
-  ret double %pow
+  %pow = call ninf float @powf(float %x, float 5.0e-01)
+  ret float %pow
 }
 
 define <2 x double> @pow_intrinsic_neghalf_fast(<2 x double> %x) {
 ; CHECK-LABEL: @pow_intrinsic_neghalf_fast(
 ; CHECK-NEXT:    [[TMP1:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> %x)
 ; CHECK-NEXT:    [[TMP2:%.*]] = fdiv fast <2 x double> <double 1.000000e+00, double 1.000000e+00>, [[TMP1]]
 ; CHECK-NEXT:    ret <2 x double> [[TMP2]]
 ;
   %pow = call fast <2 x double> @llvm.pow.v2f64(<2 x double> %x, <2 x double> <double -5.0e-01, double -5.0e-01>)
   ret <2 x double> %pow
 }
 
-define double @pow_intrinsic_neghalf_approx(double %x) {
-; CHECK-LABEL: @pow_intrinsic_neghalf_approx(
-; CHECK-NEXT:    [[POW:%.*]] = call afn double @llvm.pow.f64(double %x, double -5.000000e-01)
-; CHECK-NEXT:    ret double [[POW]]
+define <2 x float> @powf_intrinsic_neghalf_ninf(<2 x float> %x) {
+; CHECK-LABEL: @powf_intrinsic_neghalf_ninf(
+; CHECK-NEXT:    [[SQRT:%.*]] = call ninf <2 x float> @llvm.sqrt.v2f32(<2 x float> %x)
+; CHECK-NEXT:    [[TMP1:%.*]] = call ninf <2 x float> @llvm.fabs.v2f32(<2 x float> [[SQRT]])
+; CHECK-NEXT:    [[TMP2:%.*]] = fdiv ninf <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[TMP1]]
+; CHECK-NEXT:    ret <2 x float> [[TMP2]]
+;
+  %pow = call ninf <2 x float> @llvm.pow.v2f32(<2 x float> %x, <2 x float> <float -5.0e-01, float-5.0e-01>)
+  ret <2 x float> %pow
+}
+
+define float @powf_libcall_neghalf_approx(float %x) {
+; CHECK-LABEL: @powf_libcall_neghalf_approx(
+; CHECK-NEXT:    [[SQRTF:%.*]] = call afn float @sqrtf(float %x)
+; CHECK-NEXT:    [[TMP1:%.*]] = call afn float @llvm.fabs.f32(float [[SQRTF]])
+; CHECK-NEXT:    [[TMP2:%.*]] = fcmp afn oeq float %x, 0xFFF0000000000000
+; CHECK-NEXT:    [[DOTOP:%.*]] = fdiv afn float 1.000000e+00, [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = select i1 [[TMP2]], float 0.000000e+00, float [[DOTOP]]
+; CHECK-NEXT:    ret float [[TMP3]]
 ;
-  %pow = call afn double @llvm.pow.f64(double %x, double -5.0e-01)
-  ret double %pow
+  %pow = call afn float @powf(float %x, float -5.0e-01)
+  ret float %pow
 }
 
-define float @pow_libcall_neghalf_fast(float %x) {
+define double @pow_libcall_neghalf_fast(double %x) {
 ; CHECK-LABEL: @pow_libcall_neghalf_fast(
-; CHECK-NEXT:    [[SQRTF:%.*]] = call fast float @sqrtf(float %x)
-; CHECK-NEXT:    [[TMP1:%.*]] = fdiv fast float 1.000000e+00, [[SQRTF]]
-; CHECK-NEXT:    ret float [[TMP1]]
+; CHECK-NEXT:    [[SQRT:%.*]] = call fast double @sqrt(double %x)
+; CHECK-NEXT:    [[TMP1:%.*]] = fdiv fast double 1.000000e+00, [[SQRT]]
+; CHECK-NEXT:    ret double [[TMP1]]
 ;
-  %pow = call fast float @powf(float %x, float -5.0e-01)
-  ret float %pow
+  %pow = call fast double @pow(double %x, double -5.0e-01)
+  ret double %pow
 }
 
 declare double @llvm.pow.f64(double, double) #0
+declare float @llvm.pow.f32(float, float) #0
 declare <2 x double> @llvm.pow.v2f64(<2 x double>, <2 x double>) #0
+declare <2 x float> @llvm.pow.v2f32(<2 x float>, <2 x float>) #0
+declare <4 x float> @llvm.pow.v4f32(<4 x float>, <4 x float>) #0
 declare double @pow(double, double)
 declare float @powf(float, float)
 
 attributes #0 = { nounwind readnone speculatable }
 attributes #1 = { nounwind readnone }