gpuav: Add Pow to Sanitizer #11345
gpuav: Add Pow to Sanitizer #11345
Conversation
|
CI Vulkan-ValidationLayers build queued with queue ID 608789. |
|
CI Vulkan-ValidationLayers build # 22003 running. |
|
CI Vulkan-ValidationLayers build # 22003 failed. |
|
CI Vulkan-ValidationLayers build queued with queue ID 609020. |
|
CI Vulkan-ValidationLayers build # 22005 running. |
|
CI Vulkan-ValidationLayers build # 22005 passed. |
| // Casting produces artifacts in float value, need to memcpy | ||
| float x_value = 0.0f; | ||
| float y_value = 0.0f; | ||
| memcpy(&x_value, &error_record[kInstLogErrorParameterOffset_1], sizeof(float)); |
There was a problem hiding this comment.
How do you cast so that it does not work?
x_value = *(float*)(error_record + kInstLogErrorParameterOffset_1); should work
In practice this code seems fine, it gets optimised (https://godbolt.org/z/7rde8v97G) but still
There was a problem hiding this comment.
my 5c, since i worked a lot for memcpy cast, in case that’s the topic of discussion. it’s totally fine with all compilers, works as intrinsics for small types (does not generate code). It’s better than reinterpret cast - the latter violates aliasing rules. But if we start using c++ 20, then std::bit_cast is the way to go.
There was a problem hiding this comment.
basically without bit_cast I dont know if it is possible to write c++ that does not violate strict type aliasing rules when converting int to float.
| )glsl"; | ||
|
|
||
| SimpleZeroComputeTest(cs_source, SPV_SOURCE_GLSL, "SPIRV-Sanitizer-Pow"); | ||
| } |
There was a problem hiding this comment.
what about some slang tests?
4 participants
Another check for #9626