[PATCH] D62580: [OpenCL] Use long instead of long long in x86 builtins

[Karl-Johan Karlsson via Phabricator via cfe-commits]

Ka-Ka added inline comments.


================
Comment at: clang/test/CodeGen/builtins-x86.c:127-128
 
-  tmp_V2LLi = __builtin_ia32_undef128();
-  tmp_V4LLi = __builtin_ia32_undef256();
+  tmp_V2LLi = (V2LLi)__builtin_ia32_undef128();
+  tmp_V4LLi = (V4LLi)__builtin_ia32_undef256();
 
----------------
I don't like the introduced casts. Can we change the testcase to operate on the appropriate types instead?
The above two lines could probably be replaced by:
  tmp_V2d = __builtin_ia32_undef128();
  tmp_V4d = __builtin_ia32_undef256();
What do you think?



================
Comment at: clang/test/CodeGen/builtins-x86.c:228-231
+  tmp_V8s = (V8s)__builtin_ia32_packsswb128(tmp_V8s, tmp_V8s);
+  tmp_V4i = (V4i)__builtin_ia32_packssdw128(tmp_V4i, tmp_V4i);
+  tmp_V8s = (V8s)__builtin_ia32_packuswb128(tmp_V8s, tmp_V8s);
   tmp_V8s = __builtin_ia32_pmulhuw128(tmp_V8s, tmp_V8s);
----------------
same here?


================
Comment at: clang/test/CodeGen/builtins-x86.c:250
   tmp_V4s = __builtin_ia32_phsubsw(tmp_V4s, tmp_V4s);
-  tmp_V16c = __builtin_ia32_pmaddubsw128(tmp_V16c, tmp_V16c);
+  tmp_V16c = (V16c)__builtin_ia32_pmaddubsw128(tmp_V16c, tmp_V16c);
   tmp_V8c = __builtin_ia32_pmaddubsw(tmp_V8c, tmp_V8c);
----------------
and here?


================
Comment at: clang/test/CodeGen/builtins-x86.c:428
   tmp_V4i = __builtin_ia32_psradi128(tmp_V4i, tmp_i);
-  tmp_V8s = __builtin_ia32_pmaddwd128(tmp_V8s, tmp_V8s);
+  tmp_V8s = (V8s)__builtin_ia32_pmaddwd128(tmp_V8s, tmp_V8s);
   (void) __builtin_ia32_monitor(tmp_vp, tmp_Ui, tmp_Ui);
----------------
Can we change this line to?
  tmp_V4i = __builtin_ia32_pmaddwd128(tmp_V8s, tmp_V8s);



================
Comment at: clang/test/CodeGen/builtins-x86.c:432
   tmp_V16c = __builtin_ia32_lddqu(tmp_cCp);
-  tmp_V2LLi = __builtin_ia32_palignr128(tmp_V2LLi, tmp_V2LLi, imm_i);
-  tmp_V1LLi = __builtin_ia32_palignr(tmp_V1LLi, tmp_V1LLi, imm_i);
+  tmp_V2LLi = (V2LLi)__builtin_ia32_palignr128((V16c)tmp_V2LLi, (V16c)tmp_V2LLi, imm_i);
+  tmp_V1LLi = (V1LLi)__builtin_ia32_palignr((V8c)tmp_V1LLi, (V8c)tmp_V1LLi, imm_i);
----------------
Can we change this line to?
  tmp_V16c = __builtin_ia32_palignr128(tmp_V16c, tmp_V16c, imm_i);



================
Comment at: clang/test/CodeGen/builtins-x86.c:433
+  tmp_V2LLi = (V2LLi)__builtin_ia32_palignr128((V16c)tmp_V2LLi, (V16c)tmp_V2LLi, imm_i);
+  tmp_V1LLi = (V1LLi)__builtin_ia32_palignr((V8c)tmp_V1LLi, (V8c)tmp_V1LLi, imm_i);
 #ifdef USE_SSE4
----------------
Can we change this line to?
  tmp_V8c = __builtin_ia32_palignr(tmp_V8c, tmp_V8c, imm_i);



Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62580/new/

https://reviews.llvm.org/D62580