ARM: FMA is legal only if VFP4 is available.

rdar://13306723

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176212 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index b8ff5b3bd65170d59a4977925223e07f64915688..244dac5ca43e8da95791c62bbc5bb58d37bc7455 100644 (file)
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -554,6 +554,12 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::CTPOP,      MVT::v4i16, Custom);
     setOperationAction(ISD::CTPOP,      MVT::v8i16, Custom);
 
+    // NEON only has FMA instructions as of VFP4.
+    if (!Subtarget->hasVFP4()) {
+      setOperationAction(ISD::FMA, MVT::v2f32, Expand);
+      setOperationAction(ISD::FMA, MVT::v4f32, Expand);
+    }
+
     setTargetDAGCombine(ISD::INTRINSIC_VOID);
     setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
     setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);

diff --git a/test/CodeGen/ARM/2013-02-27-expand-vfma.ll b/test/CodeGen/ARM/2013-02-27-expand-vfma.ll
new file mode 100644 (file)
index 0000000..0e3bf23
--- /dev/null
+++ b/test/CodeGen/ARM/2013-02-27-expand-vfma.ll
@@ -0,0 +1,31 @@
+; RUN: llc < %s -mtriple=armv7-apple-darwin | FileCheck %s
+; RUN: llc < %s -mtriple=armv7s-apple-darwin | FileCheck %s -check-prefix=VFP4
+
+define <4 x float> @muladd(<4 x float> %a, <4 x float> %b, <4 x float> %c) nounwind {
+; CHECK: muladd:
+; CHECK: fmaf
+; CHECK: fmaf
+; CHECK: fmaf
+; CHECK: fmaf
+; CHECK-NOT: fmaf
+
+; CHECK-VFP4: vfma.f32
+  %tmp = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %b, <4 x float> %c, <4 x float> %a) #2
+  ret <4 x float> %tmp
+}
+
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) #1
+
+define <2 x float> @muladd2(<2 x float> %a, <2 x float> %b, <2 x float> %c) nounwind {
+; CHECK: muladd2:
+; CHECK: fmaf
+; CHECK: fmaf
+; CHECK-NOT: fmaf
+
+; CHECK-VFP4: vfma.f32
+  %tmp = tail call <2 x float> @llvm.fma.v2f32(<2 x float> %b, <2 x float> %c, <2 x float> %a) #2
+  ret <2 x float> %tmp
+}
+
+declare <2 x float> @llvm.fma.v2f32(<2 x float>, <2 x float>, <2 x float>) #1
+