Add pshufw instruction, fix inconsistencies with pextrw instruction.

src/x64/assembler-x64.cc

Index: src/x64/assembler-x64.cc
diff --git a/src/x64/assembler-x64.cc b/src/x64/assembler-x64.cc
index 96798bcee3971db24ca5ffd731022d9a473f9684..b366e66c2ae108136ac71a50c3e4c81e1c166643 100644
--- a/src/x64/assembler-x64.cc
+++ b/src/x64/assembler-x64.cc
@@ -834,6 +834,23 @@ void Assembler::bsfq(Register dst, const Operand& src) {
   emit_operand(dst, src);
 }
 
+void Assembler::pshufw(XMMRegister dst, XMMRegister src, uint8_t shuffle) {
+  EnsureSpace ensure_space(this);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x70);
+  emit(0xC0 | (dst.low_bits() << 3) | src.low_bits());
+  emit(shuffle);
+}
+
+void Assembler::pshufw(XMMRegister dst, const Operand& src, uint8_t shuffle) {
+  EnsureSpace ensure_space(this);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x70);
+  emit_operand(dst.code(), src);
+  emit(shuffle);
+}
 
 void Assembler::call(Label* L) {
   EnsureSpace ensure_space(this);
@@ -2871,13 +2888,15 @@ void Assembler::pinsrw(XMMRegister dst, const Operand& src, int8_t imm8) {
 }
 
 void Assembler::pextrw(Register dst, XMMRegister src, int8_t imm8) {
+  DCHECK(IsEnabled(SSE4_1));
   DCHECK(is_uint8(imm8));

[bbudge, 2017/03/22 19:17:05]

If I'm reading the docs correctly, this could be even stronger: DCHECK_GT(8,
imm8);

[gdeepti, 2017/03/22 20:45:34]

A agree that this check could be stronger, but I'm not sure if the assembler is
the right place for the check as the instructions only specifies that the count
operand is an 8-bit immediate, and only the 3 least-significant bits are used
for location. We already check this in the validate function on decode, will add
a check in the code generator before it comes to the assembler.  

[bbudge, 2017/03/22 21:09:40]

OK

   EnsureSpace ensure_space(this);
   emit(0x66);
-  emit_optional_rex_32(src, dst);
+  emit_optional_rex_32(dst, src);
   emit(0x0F);
-  emit(0xC5);
-  emit_sse_operand(src, dst);
+  emit(0x3A);
+  emit(0x15);
+  emit_sse_operand(dst, src);
   emit(imm8);
 }