[llvm-commits] [llvm] r133127 - in /llvm/trunk: lib/Target/ARM/ARMISelLowering.cpp test/CodeGen/ARM/vmul.ll

[Chad Rosier]

Author: mcrosier
Date: Wed Jun 15 20:21:54 2011
New Revision: 133127

URL: http://llvm.org/viewvc/llvm-project?rev=133127&view=rev
Log:
Revision r128665 added an optimization to make use of NEON multiplier
accumulator forwarding.  Specifically (from SVN log entry):

Distribute (A + B) * C to (A * C) + (B * C) to make use of NEON multiplier
accumulator forwarding:
vadd d3, d0, d1
vmul d3, d3, d2
=>
vmul d3, d0, d2
vmla d3, d1, d2

Make sure it catches cases where operand 1 is add/fadd/sub/fsub, which was
intended in the original revision.


Modified:
    llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
    llvm/trunk/test/CodeGen/ARM/vmul.ll

Modified: llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp?rev=133127&r1=133126&r2=133127&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Wed Jun 15 20:21:54 2011
@@ -5687,7 +5687,7 @@
   unsigned Opcode = N0.getOpcode();
   if (Opcode != ISD::ADD && Opcode != ISD::SUB &&
       Opcode != ISD::FADD && Opcode != ISD::FSUB) {
-    Opcode = N0.getOpcode();
+    Opcode = N1.getOpcode();
     if (Opcode != ISD::ADD && Opcode != ISD::SUB &&
         Opcode != ISD::FADD && Opcode != ISD::FSUB)
       return SDValue();

Modified: llvm/trunk/test/CodeGen/ARM/vmul.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/vmul.ll?rev=133127&r1=133126&r2=133127&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/ARM/vmul.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/vmul.ll Wed Jun 15 20:21:54 2011
@@ -492,3 +492,25 @@
   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
+}