Temporarily disable Hexagon tests. They are failing on OS X

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

diff --git a/test/CodeGen/ARM/vmul.ll b/test/CodeGen/ARM/vmul.ll
index 80ba9be3bd2d3dd09732699701e7a7098397b969..61d89bbae8356eec3502ac76ae8dbc603e34e5cb 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:
@@ -439,9 +439,9 @@ define <2 x i64> @vmull_extvec_u32(<2 x i32> %arg) nounwind {
 }
 
 ; rdar://9197392
-define void @distribue(i16* %dst, i8* %src, i32 %mul) nounwind {
+define void @distribute(i16* %dst, i8* %src, i32 %mul) nounwind {
 entry:
-; CHECK: distribue:
+; CHECK: distribute:
 ; CHECK: vmull.u8 [[REG1:(q[0-9]+)]], d{{.*}}, [[REG2:(d[0-9]+)]]
 ; CHECK: vmlal.u8 [[REG1]], d{{.*}}, [[REG2]]
   %0 = trunc i32 %mul to i8
@@ -466,3 +466,62 @@ 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 @distribute2(%struct.uint8x8_t* nocapture %dst, i8* %src, i32 %mul) nounwind {
+entry:
+; CHECK: distribute2
+; 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
+}
+
+define void @distribute2_commutative(%struct.uint8x8_t* nocapture %dst, i8* %src, i32 %mul) nounwind {
+entry:
+; CHECK: distribute2_commutative
+; 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> %2, %9
+  %11 = getelementptr inbounds %struct.uint8x8_t* %dst, i32 0, i32 0
+  store <8 x i8> %10, <8 x i8>* %11, align 8
+  ret void
+}
+
+; If one operand has a zero-extend and the other a sign-extend, vmull
+; cannot be used.
+define i16 @vmullWithInconsistentExtensions(<8 x i8> %vec) {
+; CHECK: vmullWithInconsistentExtensions
+; CHECK-NOT: vmull.s8
+  %1 = sext <8 x i8> %vec to <8 x i16>
+  %2 = mul <8 x i16> %1, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
+  %3 = extractelement <8 x i16> %2, i32 0
+  ret i16 %3
+}