[PATCH] D76959: [WebAssembly] Import wasm_simd128.h from Emscripten

[Dan Gohman via Phabricator via cfe-commits]

sunfish accepted this revision.
sunfish added a comment.
This revision is now accepted and ready to land.

Cool, LGTM, with optional suggestion for signed char below:



================
Comment at: clang/lib/Headers/wasm_simd128.h:30
+typedef long long __i64x2 __attribute__((__vector_size__(16), __aligned__(16)));
+typedef unsigned long long __u64x2
+    __attribute__((__vector_size__(16), __aligned__(16)));
----------------
tlively wrote:
> sunfish wrote:
> > Since this file is already including <stdint.h>, instead of `char`, `unsigned char`, `short`, `unsigned short`, etc., can these use `int8_t`, `uint8_t`, `int16_t`, `uint16_t`, and so on, for clarity? Also, this would avoid any possible behavior change under `-funsigned-char`.
> Unfortunately not :( This change gives errors like
> 
> ```
> error: cannot initialize a parameter of type '__attribute__((__vector_size__(16 * sizeof(char)))) char' (vector of 16 'char' values) with an rvalue of type '__i8x16' (vector of 16 'int8_t' values)
>   return (v128_t)__builtin_wasm_abs_i8x16((__i8x16)a);
> ```
> 
> Even changing `char` to `signed char` doesn't work. I would change the builtins to use the better types, but the builtin definition system does not allow those more interesting types.
> 
> This is especially annoying because `-funsigned-char` does indeed change the behavior of the comparison intrinsics. I will have to do extra work in those functions to make it work.
Would changing the clang definitions of the builtins to use signed char (Sc) explicitly fix this, like:

-TARGET_BUILTIN(__builtin_wasm_swizzle_v8x16, "V16cV16cV16c", "nc", "unimplemented-simd128")
+TARGET_BUILTIN(__builtin_wasm_swizzle_v8x16, "V16ScV16ScV16Sc", "nc", "unimplemented-simd128")

and so on for all the wasm simd builtins?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D76959/new/

https://reviews.llvm.org/D76959