From 5c683972bc6c5bd29bd25891b9f54d73d47bce10 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@gmail.com>
Date: Mon, 28 Jan 2013 07:41:18 +0000
Subject: [PATCH] Fix 256-bit PALIGNR comment decoding to understand that it
 works on independent 256-bit lanes.

llvm-svn: 173674
---
 llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp | 13 +++++++++++--
 llvm/test/MC/X86/shuffle-comments.s            | 15 +++++++++++++++
 2 files changed, 26 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp b/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
index b490f27..bbd4904 100644
--- a/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
+++ b/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
@@ -66,8 +66,17 @@ void DecodePALIGNRMask(MVT VT, unsigned Imm,
   unsigned NumElts = VT.getVectorNumElements();
   unsigned Offset = Imm * (VT.getVectorElementType().getSizeInBits() / 8);
 
-  for (unsigned i = 0; i != NumElts; ++i)
-    ShuffleMask.push_back((i + Offset) % (NumElts * 2));
+  unsigned NumLanes = VT.getSizeInBits() / 128;
+  unsigned NumLaneElts = NumElts / NumLanes;
+
+  for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
+    for (unsigned i = 0; i != NumLaneElts; ++i) {
+      unsigned Base = i + Offset;
+      // if i+offset is out of this lane then we actually need the other source
+      if (Base >= NumLaneElts) Base += NumElts - NumLaneElts;
+      ShuffleMask.push_back(Base + l);
+    }
+  }
 }
 
 /// DecodePSHUFMask - This decodes the shuffle masks for pshufd, and vpermilp*.
diff --git a/llvm/test/MC/X86/shuffle-comments.s b/llvm/test/MC/X86/shuffle-comments.s
index e2f84af..d38fe03 100644
--- a/llvm/test/MC/X86/shuffle-comments.s
+++ b/llvm/test/MC/X86/shuffle-comments.s
@@ -29,3 +29,18 @@ vpalignr $0, %xmm0, %xmm1, %xmm2
 # CHECK: xmm2 = xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
 vpalignr $0, (%rax), %xmm1, %xmm2
 # CHECK: xmm2 = mem[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
+
+vpalignr $8, %ymm0, %ymm1, %ymm2
+# CHECK: ymm2 = ymm0[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm0[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+vpalignr $8, (%rax), %ymm1, %ymm2
+# CHECK: ymm2 = mem[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],mem[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+
+vpalignr $16, %ymm0, %ymm1, %ymm2
+# CHECK: ymm2 = ymm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31]
+vpalignr $16, (%rax), %ymm1, %ymm2
+# CHECK: ymm2 = ymm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31]
+
+vpalignr $0, %ymm0, %ymm1, %ymm2
+# CHECK: ymm2 = ymm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31]
+vpalignr $0, (%rax), %ymm1, %ymm2
+# CHECK: ymm2 = mem[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31]
-- 
2.7.4