Details
Diff Detail
Event Timeline
lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp | ||
---|---|---|
286 | This should be able to handle any bitwidth < 32. We will want this for the < i16 types that will want to be legalized to i16 as well | |
293–296 | You should just define these when declared instead of resetting them below | |
424–425 | This should be a switch since other intrinsics are likely, and it can just return the result of visitBitreverseIntrinsicInst directly | |
test/CodeGen/AMDGPU/amdgpu-codegenprepare-i16-to-i32.ll | ||
509 | Should have a test with smaller bitwidth | |
510–515 | There should be check lines for the non-VI case for the unprompted case |
lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp | ||
---|---|---|
286 | I am working on handling smaller types in instructions and bitereverse in a separate patch, it is almost ready (I need to clean up and update doxygen). smaller types patch fixes a bunch of test failures on Wei's fp16 branch. | |
test/CodeGen/AMDGPU/amdgpu-codegenprepare-i16-to-i32.ll | ||
509 | In a separate patch. |
This should be able to handle any bitwidth < 32. We will want this for the < i16 types that will want to be legalized to i16 as well