[wasm-simd] Move more macro-assembler functions into SharedTurboAssembler

Bug: v8:11589
Change-Id: I875135f4a39f56f14a1f08e149a70b1d884bf68a
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2808005Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73816}

src/codegen/ia32/macro-assembler-ia32.h

@@ -468,52 +468,6 @@ class V8_EXPORT_PRIVATE TurboAssembler : public SharedTurboAssembler {
   AVX_OP3_WITH_MOVE(Pmaddwd, pmaddwd, XMMRegister, Operand)
 #undef AVX_OP3_WITH_MOVE
 
-// Non-SSE2 instructions.
-#define AVX_OP2_WITH_TYPE_SCOPE(macro_name, name, dst_type, src_type, \
-                                sse_scope)                            \
-  void macro_name(dst_type dst, src_type src) {                       \
-    if (CpuFeatures::IsSupported(AVX)) {                              \
-      CpuFeatureScope scope(this, AVX);                               \
-      v##name(dst, src);                                              \
-      return;                                                         \
-    }                                                                 \
-    if (CpuFeatures::IsSupported(sse_scope)) {                        \
-      CpuFeatureScope scope(this, sse_scope);                         \
-      name(dst, src);                                                 \
-      return;                                                         \
-    }                                                                 \
-    UNREACHABLE();                                                    \
-  }
-#define AVX_OP2_XO_SSE3(macro_name, name)                                   \
-  AVX_OP2_WITH_TYPE_SCOPE(macro_name, name, XMMRegister, XMMRegister, SSE3) \
-  AVX_OP2_WITH_TYPE_SCOPE(macro_name, name, XMMRegister, Operand, SSE3)
-  AVX_OP2_XO_SSE3(Movddup, movddup)
-  AVX_OP2_WITH_TYPE_SCOPE(Movshdup, movshdup, XMMRegister, XMMRegister, SSE3)
-
-#undef AVX_OP2_XO_SSE3
-
-#define AVX_OP2_XO_SSSE3(macro_name, name)                                   \
-  AVX_OP2_WITH_TYPE_SCOPE(macro_name, name, XMMRegister, XMMRegister, SSSE3) \
-  AVX_OP2_WITH_TYPE_SCOPE(macro_name, name, XMMRegister, Operand, SSSE3)
-  AVX_OP2_XO_SSSE3(Pabsb, pabsb)
-  AVX_OP2_XO_SSSE3(Pabsw, pabsw)
-  AVX_OP2_XO_SSSE3(Pabsd, pabsd)
-
-#undef AVX_OP2_XO_SSE3
-
-#define AVX_OP2_XO_SSE4(macro_name, name)                                     \
-  AVX_OP2_WITH_TYPE_SCOPE(macro_name, name, XMMRegister, XMMRegister, SSE4_1) \
-  AVX_OP2_WITH_TYPE_SCOPE(macro_name, name, XMMRegister, Operand, SSE4_1)
-
-  AVX_OP2_XO_SSE4(Ptest, ptest)
-  AVX_OP2_XO_SSE4(Pmovsxwd, pmovsxwd)
-  AVX_OP2_XO_SSE4(Pmovsxdq, pmovsxdq)
-  AVX_OP2_XO_SSE4(Pmovzxwd, pmovzxwd)
-  AVX_OP2_XO_SSE4(Pmovzxdq, pmovzxdq)
-
-#undef AVX_OP2_WITH_TYPE_SCOPE
-#undef AVX_OP2_XO_SSE4
-
 #define AVX_OP3_WITH_TYPE_SCOPE(macro_name, name, dst_type, src_type, \
                                 sse_scope)                            \
   void macro_name(dst_type dst, dst_type src1, src_type src2) {       \

src/codegen/shared-ia32-x64/macro-assembler-shared-ia32-x64.h

@@ -136,11 +136,21 @@ class V8_EXPORT_PRIVATE SharedTurboAssembler : public TurboAssemblerBase {
   AVX_OP(Rsqrtps, rsqrtps)
   AVX_OP(Sqrtps, sqrtps)
   AVX_OP(Sqrtpd, sqrtpd)
+  AVX_OP_SSE3(Movddup, movddup)
+  AVX_OP_SSE3(Movshdup, movshdup)
+  AVX_OP_SSSE3(Pabsb, pabsb)
+  AVX_OP_SSSE3(Pabsw, pabsw)
+  AVX_OP_SSSE3(Pabsd, pabsd)
   AVX_OP_SSE4_1(Extractps, extractps)
   AVX_OP_SSE4_1(Pextrb, pextrb)
   AVX_OP_SSE4_1(Pextrw, pextrw)
   AVX_OP_SSE4_1(Pmovsxbw, pmovsxbw)
+  AVX_OP_SSE4_1(Pmovsxwd, pmovsxwd)
+  AVX_OP_SSE4_1(Pmovsxdq, pmovsxdq)
   AVX_OP_SSE4_1(Pmovzxbw, pmovzxbw)
+  AVX_OP_SSE4_1(Pmovzxwd, pmovzxwd)
+  AVX_OP_SSE4_1(Pmovzxdq, pmovzxdq)
+  AVX_OP_SSE4_1(Ptest, ptest)
   AVX_OP_SSE4_1(Roundps, roundps)
   AVX_OP_SSE4_1(Roundpd, roundpd)
 

src/codegen/x64/macro-assembler-x64.h

@@ -162,8 +162,6 @@ class V8_EXPORT_PRIVATE TurboAssembler : public SharedTurboAssembler {
   AVX_OP(Cmppd, cmppd)
   AVX_OP(Movlhps, movlhps)
   AVX_OP_SSE3(Haddps, haddps)
-  AVX_OP_SSE3(Movddup, movddup)
-  AVX_OP_SSE3(Movshdup, movshdup)
   AVX_OP_SSSE3(Phaddd, phaddd)
   AVX_OP_SSSE3(Phaddw, phaddw)
   AVX_OP_SSSE3(Pshufb, pshufb)
@@ -171,9 +169,6 @@ class V8_EXPORT_PRIVATE TurboAssembler : public SharedTurboAssembler {
   AVX_OP_SSSE3(Psignw, psignw)
   AVX_OP_SSSE3(Psignd, psignd)
   AVX_OP_SSSE3(Palignr, palignr)
-  AVX_OP_SSSE3(Pabsb, pabsb)
-  AVX_OP_SSSE3(Pabsw, pabsw)
-  AVX_OP_SSSE3(Pabsd, pabsd)
   AVX_OP_SSE4_1(Pcmpeqq, pcmpeqq)
   AVX_OP_SSE4_1(Packusdw, packusdw)
   AVX_OP_SSE4_1(Pminsb, pminsb)
@@ -188,11 +183,6 @@ class V8_EXPORT_PRIVATE TurboAssembler : public SharedTurboAssembler {
   AVX_OP_SSE4_1(Insertps, insertps)
   AVX_OP_SSE4_1(Pinsrq, pinsrq)
   AVX_OP_SSE4_1(Pblendw, pblendw)
-  AVX_OP_SSE4_1(Ptest, ptest)
-  AVX_OP_SSE4_1(Pmovsxwd, pmovsxwd)
-  AVX_OP_SSE4_1(Pmovsxdq, pmovsxdq)
-  AVX_OP_SSE4_1(Pmovzxwd, pmovzxwd)
-  AVX_OP_SSE4_1(Pmovzxdq, pmovzxdq)
   AVX_OP_SSE4_1(Pextrq, pextrq)
   AVX_OP_SSE4_1(Roundss, roundss)
   AVX_OP_SSE4_1(Roundsd, roundsd)