[PPC64LE] Generate correct little-endian code for v16i8 multiply

The existing code in PPCTargetLowering::LowerMUL() for multiplying two
v16i8 values assumes that vector elements are numbered in big-endian
order.  For little-endian targets, the vector element numbering is
reversed, but the vmuleub, vmuloub, and vperm instructions still
assume big-endian numbering.  To account for this, we must adjust the
permute control vector and reverse the order of the input registers on
the vperm instruction.

The existing test/CodeGen/PowerPC/vec_mul.ll is updated to be executed
on powerpc64 and powerpc64le targets as well as the original powerpc
(32-bit) target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210474 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index d5a685d9f482f72c5b73b1fedbbbcee860d70a8d..91366e2ccebfbbb3bc2e0f44fa41d72cdfc0940f 100644 (file)
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -6059,6 +6059,7 @@ SDValue PPCTargetLowering::LowerMUL(SDValue Op, SelectionDAG &DAG) const {
                             LHS, RHS, Zero, DAG, dl);
   } else if (Op.getValueType() == MVT::v16i8) {
     SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
+    bool isLittleEndian = PPCSubTarget.isLittleEndian();
 
     // Multiply the even 8-bit parts, producing 16-bit sums.
     SDValue EvenParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuleub,
@@ -6070,13 +6071,24 @@ SDValue PPCTargetLowering::LowerMUL(SDValue Op, SelectionDAG &DAG) const {
                                           LHS, RHS, DAG, dl, MVT::v8i16);
     OddParts = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, OddParts);
 
-    // Merge the results together.
+    // Merge the results together.  Because vmuleub and vmuloub are
+    // instructions with a big-endian bias, we must reverse the
+    // element numbering and reverse the meaning of "odd" and "even"
+    // when generating little endian code.
     int Ops[16];
     for (unsigned i = 0; i != 8; ++i) {
-      Ops[i*2  ] = 2*i+1;
-      Ops[i*2+1] = 2*i+1+16;
+      if (isLittleEndian) {
+        Ops[i*2  ] = 2*i;
+        Ops[i*2+1] = 2*i+16;
+      } else {
+        Ops[i*2  ] = 2*i+1;
+        Ops[i*2+1] = 2*i+1+16;
+      }
     }
-    return DAG.getVectorShuffle(MVT::v16i8, dl, EvenParts, OddParts, Ops);
+    if (isLittleEndian)
+      return DAG.getVectorShuffle(MVT::v16i8, dl, OddParts, EvenParts, Ops);
+    else
+      return DAG.getVectorShuffle(MVT::v16i8, dl, EvenParts, OddParts, Ops);
   } else {
     llvm_unreachable("Unknown mul to lower!");
   }

diff --git a/test/CodeGen/PowerPC/vec_mul.ll b/test/CodeGen/PowerPC/vec_mul.ll
index c376751d80609e6594a564b01c018f4ed6e74755..8a448156c98e847806b3cdeab622bf6f2fb74cd7 100644 (file)
--- a/test/CodeGen/PowerPC/vec_mul.ll
+++ b/test/CodeGen/PowerPC/vec_mul.ll
@@ -1,4 +1,6 @@
 ; RUN: llc < %s -mtriple=powerpc-unknown-linux-gnu -march=ppc32 -mattr=+altivec | FileCheck %s
+; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -march=ppc64 -mattr=+altivec | FileCheck %s
+; RUN: llc < %s -mtriple=powerpc64le-unknown-linux-gnu -march=ppc64 -mattr=+altivec | FileCheck %s -check-prefix=CHECK-LE
 
 define <4 x i32> @test_v4i32(<4 x i32>* %X, <4 x i32>* %Y) {
        %tmp = load <4 x i32>* %X               ; <<4 x i32>> [#uses=1]
@@ -9,6 +11,9 @@ define <4 x i32> @test_v4i32(<4 x i32>* %X, <4 x i32>* %Y) {
 ; CHECK-LABEL: test_v4i32:
 ; CHECK: vmsumuhm
 ; CHECK-NOT: mullw
+; CHECK-LE-LABEL: test_v4i32:
+; CHECK-LE: vmsumuhm
+; CHECK-LE-NOT: mullw
 
 define <8 x i16> @test_v8i16(<8 x i16>* %X, <8 x i16>* %Y) {
        %tmp = load <8 x i16>* %X               ; <<8 x i16>> [#uses=1]
@@ -19,6 +24,9 @@ define <8 x i16> @test_v8i16(<8 x i16>* %X, <8 x i16>* %Y) {
 ; CHECK-LABEL: test_v8i16:
 ; CHECK: vmladduhm
 ; CHECK-NOT: mullw
+; CHECK-LE-LABEL: test_v8i16:
+; CHECK-LE: vmladduhm
+; CHECK-LE-NOT: mullw
 
 define <16 x i8> @test_v16i8(<16 x i8>* %X, <16 x i8>* %Y) {
        %tmp = load <16 x i8>* %X               ; <<16 x i8>> [#uses=1]
@@ -30,6 +38,11 @@ define <16 x i8> @test_v16i8(<16 x i8>* %X, <16 x i8>* %Y) {
 ; CHECK: vmuloub
 ; CHECK: vmuleub
 ; CHECK-NOT: mullw
+; CHECK-LE-LABEL: test_v16i8:
+; CHECK-LE: vmuloub [[REG1:[0-9]+]]
+; CHECK-LE: vmuleub [[REG2:[0-9]+]]
+; CHECK-LE: vperm {{[0-9]+}}, [[REG2]], [[REG1]]
+; CHECK-LE-NOT: mullw
 
 define <4 x float> @test_float(<4 x float>* %X, <4 x float>* %Y) {
        %tmp = load <4 x float>* %X
@@ -44,3 +57,7 @@ define <4 x float> @test_float(<4 x float>* %X, <4 x float>* %Y) {
 ; CHECK: vspltisw [[ZNEG:[0-9]+]], -1
 ; CHECK: vslw     {{[0-9]+}}, [[ZNEG]], [[ZNEG]]
 ; CHECK: vmaddfp
+; CHECK-LE-LABEL: test_float:
+; CHECK-LE: vspltisw [[ZNEG:[0-9]+]], -1
+; CHECK-LE: vslw     {{[0-9]+}}, [[ZNEG]], [[ZNEG]]
+; CHECK-LE: vmaddfp