Implement MUBUF instructions for shorts/bytes (#2856)

* implement loads/store instructions for types smaller than dwords

* initialize s16/s8 types

* set profile for int8/16/64

* also need to zero extend u8/u16 to u32 result

* document unrelated bugs with atomic fmin/max

* remove profile checks and simple emit for added opcodes

---------

Co-authored-by: georgemoralis <giorgosmrls@gmail.com>
This commit is contained in:
baggins183
2025-07-18 02:04:50 -07:00
committed by GitHub
parent 76f003d388
commit 3019bfb978
11 changed files with 120 additions and 58 deletions

View File

@@ -488,6 +488,8 @@ Id EmitConvertU16U32(EmitContext& ctx, Id value);
Id EmitConvertU32U16(EmitContext& ctx, Id value);
Id EmitConvertU8U32(EmitContext& ctx, Id value);
Id EmitConvertU32U8(EmitContext& ctx, Id value);
Id EmitConvertS32S8(EmitContext& ctx, Id value);
Id EmitConvertS32S16(EmitContext& ctx, Id value);
Id EmitImageSampleRaw(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address1, Id address2,
Id address3, Id address4);