Don't pass FPOpFusion::Strict to the backend

[anemet]

Summary

This restores the behavior prior to D31167 where the code-gen default was
FPC_On which mapped to FPOpFusion::Standard. After merging the FE
state (on/off) and the code-gen state (on/fast/off), the default became off to
match the front-end.

In other words, the front-end controls when to fuse along the language
standards and the backend shouldn't override this by splitting fused
intrinsics as FPOpFusion::Strict would imply.

[hfinkel]

LGTM

lib/CodeGen/BackendUtil.cpp

Options.FloatABIType =
           .Case("soft", llvm::FloatABI::Soft)
           .Case("softfp", llvm::FloatABI::Soft)
           .Case("hard", llvm::FloatABI::Hard)
           .Default(llvm::FloatABI::Default);
 
   // Set FP fusion mode.
   switch (LangOpts.getDefaultFPContractMode()) {
   case LangOptions::FPC_Off:
-    Options.AllowFPOpFusion = llvm::FPOpFusion::Strict;
+    // Preserve any contraction performed by the front-end.  (Strict performs
+    // splitting of the muladd instrinsic in the backend.)
+    Options.AllowFPOpFusion = llvm::FPOpFusion::Standard;
     break;
   case LangOptions::FPC_On:
     Options.AllowFPOpFusion = llvm::FPOpFusion::Standard;
     break;
   case LangOptions::FPC_Fast:
     Options.AllowFPOpFusion = llvm::FPOpFusion::Fast;
     break;
   }

test/CodeGen/fp-contract-on-asm.c

+// RUN: %clang_cc1 -O3 -triple=aarch64-apple-ios -S -o - %s | FileCheck %s
+// REQUIRES: aarch64-registered-target
+
+float fma_test1(float a, float b, float c) {
+#pragma STDC FP_CONTRACT ON
+// CHECK-LABEL: fma_test1:
+// CHECK: fmadd
+  float x = a * b + c;
+  return x;
+}
+
+float fma_test2(float a, float b, float c) {
+// CHECK-LABEL: fma_test2:
+// CHECK: fmul
+// CHECK: fadd
+  float x = a * b + c;
+  return x;
+}