Distribute (A + B) * C to (A * C) + (B * C) to make use of NEON multiplier

[oota-llvm.git] / test / CodeGen / ARM / vmul.ll

diff --git a/test/CodeGen/ARM/vmul.ll b/test/CodeGen/ARM/vmul.ll
index 80ba9be3bd2d3dd09732699701e7a7098397b969..1fd6581ae0816b1bf5a0e426fd8ffa645fdc11f6 100644 (file)
--- a/test/CodeGen/ARM/vmul.ll
+++ b/test/CodeGen/ARM/vmul.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
+; RUN: llc < %s -march=arm -mcpu=cortex-a8 | FileCheck %s
 
 define <8 x i8> @vmuli8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
 ;CHECK: vmuli8:
@@ -466,3 +466,29 @@ entry:
 declare <16 x i8> @llvm.arm.neon.vld1.v16i8(i8*, i32) nounwind readonly
 
 declare void @llvm.arm.neon.vst1.v8i16(i8*, <8 x i16>, i32) nounwind
+
+; Take advantage of the Cortex-A8 multiplier accumulator forward.
+
+%struct.uint8x8_t = type { <8 x i8> }
+
+define void @distribue2(%struct.uint8x8_t* nocapture %dst, i8* %src, i32 %mul) nounwind {
+entry:
+; CHECK: distribue2
+; CHECK-NOT: vadd.i8
+; CHECK: vmul.i8
+; CHECK: vmla.i8
+  %0 = trunc i32 %mul to i8
+  %1 = insertelement <8 x i8> undef, i8 %0, i32 0
+  %2 = shufflevector <8 x i8> %1, <8 x i8> undef, <8 x i32> zeroinitializer
+  %3 = tail call <16 x i8> @llvm.arm.neon.vld1.v16i8(i8* %src, i32 1)
+  %4 = bitcast <16 x i8> %3 to <2 x double>
+  %5 = extractelement <2 x double> %4, i32 1
+  %6 = bitcast double %5 to <8 x i8>
+  %7 = extractelement <2 x double> %4, i32 0
+  %8 = bitcast double %7 to <8 x i8>
+  %9 = add <8 x i8> %6, %8
+  %10 = mul <8 x i8> %9, %2
+  %11 = getelementptr inbounds %struct.uint8x8_t* %dst, i32 0, i32 0
+  store <8 x i8> %10, <8 x i8>* %11, align 8
+  ret void
+}