From: Quinn Pham <quinn.pham@ibm.com>
Date: Tue, 20 Jul 2021 13:57:18 +0000 (-0500)
Subject: [PowerPC] Restore FastMathFlags of Builder for Vector FDiv Builtins
X-Git-Tag: llvmorg-14-init~903
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=fd855c24c72ce01573d726317acaaefc9809d9dc;p=platform%2Fupstream%2Fllvm.git

[PowerPC] Restore FastMathFlags of Builder for Vector FDiv Builtins

This patch fixes `__builtin_ppc_recipdivf`, `__builtin_ppc_recipdivd`,
`__builtin_ppc_rsqrtf`, and `__builtin_ppc_rsqrtd`. FastMathFlags are
set to fast immediately before emitting these builtins. Now the flags
are restored to their previous values after the builtins are emitted.

Reviewed By: nemanjai, #powerpc

Differential Revision: https://reviews.llvm.org/D105984
---

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index d9958fe..44aabfd 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -15280,6 +15280,7 @@ Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID,
   case PPC::BI__builtin_ppc_recipdivd:
   case PPC::BI__builtin_ppc_rsqrtf:
   case PPC::BI__builtin_ppc_rsqrtd: {
+    FastMathFlags FMF = Builder.getFastMathFlags();
     Builder.getFastMathFlags().setFast();
     llvm::Type *ResultType = ConvertType(E->getType());
     Value *X = EmitScalarExpr(E->getArg(0));
@@ -15287,11 +15288,15 @@ Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID,
     if (BuiltinID == PPC::BI__builtin_ppc_recipdivf ||
         BuiltinID == PPC::BI__builtin_ppc_recipdivd) {
       Value *Y = EmitScalarExpr(E->getArg(1));
-      return Builder.CreateFDiv(X, Y, "recipdiv");
+      Value *FDiv = Builder.CreateFDiv(X, Y, "recipdiv");
+      Builder.getFastMathFlags() &= (FMF);
+      return FDiv;
     }
     auto *One = ConstantFP::get(ResultType, 1.0);
     llvm::Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType);
-    return Builder.CreateFDiv(One, Builder.CreateCall(F, X), "rsqrt");
+    Value *FDiv = Builder.CreateFDiv(One, Builder.CreateCall(F, X), "rsqrt");
+    Builder.getFastMathFlags() &= (FMF);
+    return FDiv;
   }
   case PPC::BI__builtin_ppc_alignx: {
     ConstantInt *AlignmentCI = cast<ConstantInt>(Ops[0]);
diff --git a/clang/test/CodeGen/builtins-ppc-fastmath.c b/clang/test/CodeGen/builtins-ppc-fastmath.c
new file mode 100644
index 0000000..5e7aa55
--- /dev/null
+++ b/clang/test/CodeGen/builtins-ppc-fastmath.c
@@ -0,0 +1,70 @@
+// REQUIRES: powerpc-registered-target
+// RUN: %clang_cc1 -triple powerpc64-unknown-unknown \
+// RUN:   -emit-llvm %s -o - -target-cpu pwr7 | FileCheck %s
+// RUN: %clang_cc1 -triple powerpc64le-unknown-unknown \
+// RUN:   -emit-llvm %s -o - -target-cpu pwr8 | FileCheck %s
+// RUN: %clang_cc1 -triple powerpc64-unknown-aix \
+// RUN:   -emit-llvm %s -o - -target-cpu pwr7 | FileCheck %s
+// RUN: %clang_cc1 -triple powerpc-unknown-aix \
+// RUN:   -emit-llvm %s -o - -target-cpu pwr7 | FileCheck %s
+
+extern vector float a;
+extern vector float b;
+extern vector float c;
+extern vector double d;
+extern vector double e;
+extern vector double f;
+
+// CHECK-LABEL: @test_flags_recipdivf(
+// CHECK:    [[TMP0:%.*]] = load <4 x float>, <4 x float>* @a, align 16
+// CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* @b, align 16
+// CHECK-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* @a, align 16
+// CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* @b, align 16
+// CHECK-NEXT:    [[RECIPDIV:%.*]] = fdiv fast <4 x float> [[TMP2]], [[TMP3]]
+// CHECK-NEXT:    [[TMP4:%.*]] = load <4 x float>, <4 x float>* @c, align 16
+// CHECK-NEXT:    [[ADD:%.*]] = fadd <4 x float> [[RECIPDIV]], [[TMP4]]
+// CHECK-NEXT:    ret <4 x float> [[ADD]]
+//
+vector float test_flags_recipdivf() {
+  return __builtin_ppc_recipdivf(a, b) + c;
+}
+
+// CHECK-LABEL: @test_flags_recipdivd(
+// CHECK:    [[TMP0:%.*]] = load <2 x double>, <2 x double>* @d, align 16
+// CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* @e, align 16
+// CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* @d, align 16
+// CHECK-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* @e, align 16
+// CHECK-NEXT:    [[RECIPDIV:%.*]] = fdiv fast <2 x double> [[TMP2]], [[TMP3]]
+// CHECK-NEXT:    [[TMP4:%.*]] = load <2 x double>, <2 x double>* @f, align 16
+// CHECK-NEXT:    [[ADD:%.*]] = fadd <2 x double> [[RECIPDIV]], [[TMP4]]
+// CHECK-NEXT:    ret <2 x double> [[ADD]]
+//
+vector double test_flags_recipdivd() {
+  return __builtin_ppc_recipdivd(d, e) + f;
+}
+
+// CHECK-LABEL: @test_flags_rsqrtf(
+// CHECK:    [[TMP0:%.*]] = load <4 x float>, <4 x float>* @a, align 16
+// CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* @a, align 16
+// CHECK-NEXT:    [[TMP2:%.*]] = call fast <4 x float> @llvm.sqrt.v4f32(<4 x float> [[TMP1]])
+// CHECK-NEXT:    [[RSQRT:%.*]] = fdiv fast <4 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>, [[TMP2]]
+// CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* @b, align 16
+// CHECK-NEXT:    [[ADD:%.*]] = fadd <4 x float> [[RSQRT]], [[TMP3]]
+// CHECK-NEXT:    ret <4 x float> [[ADD]]
+//
+vector float test_flags_rsqrtf() {
+  return __builtin_ppc_rsqrtf(a) + b;
+}
+
+// CHECK-LABEL: @test_flags_rsqrtd(
+// CHECK:    [[TMP0:%.*]] = load <2 x double>, <2 x double>* @d, align 16
+// CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* @d, align 16
+// CHECK-NEXT:    [[TMP2:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[TMP1]])
+// CHECK-NEXT:    [[RSQRT:%.*]] = fdiv fast <2 x double> <double 1.000000e+00, double 1.000000e+00>, [[TMP2]]
+// CHECK-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* @e, align 16
+// CHECK-NEXT:    [[ADD:%.*]] = fadd <2 x double> [[RSQRT]], [[TMP3]]
+// CHECK-NEXT:    ret <2 x double> [[ADD]]
+//
+vector double test_flags_rsqrtd() {
+  return __builtin_ppc_rsqrtd(d) + e;
+}