From: Bruce Forstall <brucefo@microsoft.com>
Date: Sat, 27 Aug 2016 01:25:42 +0000 (-0700)
Subject: RyuJIT x86: Fix initblk with unrolled loop and constant fill
X-Git-Tag: submit/tizen/20210909.063632~11030^2~9553^2
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=d67a4751f3bfe7b2874abef3db2ce0975707f14f;p=platform%2Fupstream%2Fdotnet%2Fruntime.git

RyuJIT x86: Fix initblk with unrolled loop and constant fill

The computed fill value was only using the low 8 bytes of the
xmm register; we need it to fill the entire 16 byte register.

Fixes dotnet/coreclr#6940


Commit migrated from https://github.com/dotnet/coreclr/commit/5127eef602e0c0ed5ab7f63a9bc28b81558b2084
---

diff --git a/src/coreclr/src/jit/codegenxarch.cpp b/src/coreclr/src/jit/codegenxarch.cpp
index 5e35067..e616305 100755
--- a/src/coreclr/src/jit/codegenxarch.cpp
+++ b/src/coreclr/src/jit/codegenxarch.cpp
@@ -3489,8 +3489,12 @@ void CodeGen::genCodeForInitBlkUnroll(GenTreeInitBlk* initBlkNode)
 
         if (initVal->gtIntCon.gtIconVal != 0)
         {
-            emit->emitIns_R_R(INS_mov_i2xmm, EA_8BYTE, tmpReg, valReg);
+            emit->emitIns_R_R(INS_mov_i2xmm, EA_PTRSIZE, tmpReg, valReg);
             emit->emitIns_R_R(INS_punpckldq, EA_8BYTE, tmpReg, tmpReg);
+#ifdef _TARGET_X86_
+            // For x86, we need one more to convert it from 8 bytes to 16 bytes.
+            emit->emitIns_R_R(INS_punpckldq, EA_8BYTE, tmpReg, tmpReg);
+#endif // _TARGET_X86_
         }
         else
         {