[PATCH] D51435: [SLC] Support expanding pow(x, n+0.5) to x * x * ... * sqrt(x)

[Florian Hahn via Phabricator via llvm-commits]

fhahn created this revision.
fhahn added reviewers: evandro, efriedma, spatel.

https://reviews.llvm.org/D51435

Files:
  lib/Transforms/Utils/SimplifyLibCalls.cpp
  test/Transforms/InstCombine/pow-4.ll


Index: test/Transforms/InstCombine/pow-4.ll
===================================================================
--- test/Transforms/InstCombine/pow-4.ll
+++ test/Transforms/InstCombine/pow-4.ll
@@ -117,3 +117,33 @@
   ret double %1
 }
 
+; pow(x, 16.5)
+define double @test_simplify_16_5(double %x) {
+; CHECK-LABEL: @test_simplify_16_5(
+; CHECK-NEXT:    [[SQUARE:%.*]] = fmul fast double [[X:%.*]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul fast double [[SQUARE]], [[SQUARE]]
+; CHECK-NEXT:    [[TMP2:%.*]] = fmul fast double [[TMP1]], [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = fmul fast double [[TMP2]], [[TMP2]]
+; CHECK-NEXT:    [[SQRT:%.*]] = call fast double @sqrt(double [[X]]) #1
+; CHECK-NEXT:    [[TMP4:%.*]] = fmul fast double [[TMP3]], [[SQRT]]
+; CHECK-NEXT:    ret double [[TMP4]]
+;
+  %1 = call fast double @llvm.pow.f64(double %x, double 1.650000e+01)
+  ret double %1
+}
+
+; pow(x, -16.5)
+define double @test_simplify_neg_16_5(double %x) {
+; CHECK-LABEL: @test_simplify_neg_16_5(
+; CHECK-NEXT:    [[SQUARE:%.*]] = fmul fast double [[X:%.*]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul fast double [[SQUARE]], [[SQUARE]]
+; CHECK-NEXT:    [[TMP2:%.*]] = fmul fast double [[TMP1]], [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = fmul fast double [[TMP2]], [[TMP2]]
+; CHECK-NEXT:    [[SQRT:%.*]] = call fast double @sqrt(double [[X]]) #1
+; CHECK-NEXT:    [[TMP4:%.*]] = fmul fast double [[TMP3]], [[SQRT]]
+; CHECK-NEXT:    [[RECIPROCAL:%.*]] = fdiv fast double 1.000000e+00, [[TMP4]]
+; CHECK-NEXT:    ret double [[RECIPROCAL]]
+;
+  %1 = call fast double @llvm.pow.f64(double %x, double -1.650000e+01)
+  ret double %1
+}
Index: lib/Transforms/Utils/SimplifyLibCalls.cpp
===================================================================
--- lib/Transforms/Utils/SimplifyLibCalls.cpp
+++ lib/Transforms/Utils/SimplifyLibCalls.cpp
@@ -1354,7 +1354,22 @@
     // 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) {
+    if (ExpoA.compare(LimF) == APFloat::cmpLessThan) {
+      // This transformation applies to integer or integer+0.5 exponents only.
+      bool ExpoIsIntPlusHalf = false;
+      if (!ExpoA.isInteger()) {
+        APFloat VInt = ExpoA;
+        if (VInt.add(ExpoA, APFloat::rmNearestTiesToEven) != APFloat::opOK)
+          return nullptr;
+
+        bool HasSqrtFn = hasUnaryFloatFn(TLI, Ty, LibFunc_sqrt,
+                                         LibFunc_sqrtf, LibFunc_sqrtl);
+        if (!VInt.isInteger() || !HasSqrtFn)
+          return nullptr;
+
+        ExpoIsIntPlusHalf = true;
+      }
+
       // We will memoize intermediate products of the Addition Chain.
       Value *InnerChain[33] = {nullptr};
       InnerChain[1] = Base;
@@ -1365,6 +1380,13 @@
       ExpoA.convert(APFloat::IEEEdouble(), APFloat::rmTowardZero, &Ignored);
       Value *FMul = getPow(InnerChain, ExpoA.convertToDouble(), B);
 
+      // Expand pow(x, y+0.5) to pow(x, y) * sqrt(x).
+      if (ExpoIsIntPlusHalf) {
+        Value *Sqrt = emitUnaryFloatFnCall(Base, "sqrt", B,
+                                           Callee->getAttributes());
+        FMul = B.CreateFMul(FMul, Sqrt);
+      }
+
       // If the exponent is negative, then get the reciprocal.
       if (ExpoF->isNegative())
         FMul = B.CreateFDiv(ConstantFP::get(Ty, 1.0), FMul, "reciprocal");


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D51435.163106.patch
Type: text/x-patch
Size: 3476 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180829/9e2819d0/attachment.bin>