Fix issues found on linux with Vulkan GPU timers #40
Conversation
| return; | ||
|
|
||
| vkGetQueryPoolResults(S.pGPU->Devices[nNode], S.pGPU->QueryPool[nNode], nBegin, nCount, 8*nCount, &S.pGPU->nResults[nBegin], 8, VK_QUERY_RESULT_64_BIT|VK_QUERY_RESULT_PARTIAL_BIT ); | ||
| vkGetQueryPoolResults(S.pGPU->Devices[nNode], S.pGPU->QueryPool[nNode], nBegin, nCount, 8*nCount, &S.pGPU->nResults[nBegin], 8, VK_QUERY_RESULT_64_BIT); |
There was a problem hiding this comment.
https://www.khronos.org/registry/vulkan/specs/1.1-extensions/man/html/vkGetQueryPoolResults.html
VK_QUERY_RESULT_PARTIAL_BIT must not be used if the pool’s queryType is VK_QUERY_TYPE_TIMESTAMP.
maybe VK_QUERY_RESULT_WITH_AVAILABILITY_BIT is the intended behavior here? Without it we skip not-ready values, with it we write in 0 for not-ready values: I don't know enough about the surrounding code to say either way.
|
|
||
| void MicroProfileGpuShutdown() | ||
| { | ||
| VkAllocationCallbacks* noAlloc = nullptr; |
There was a problem hiding this comment.
Implementing cleanup. without this my particular laptop will crash on shutdown
| { | ||
| const uint32_t nLim = pTable->nLim; | ||
| uint32_t B = H % pTable->nAllocated; | ||
| uint32_t B = static_cast<uint32_t>(H) % pTable->nAllocated; |
There was a problem hiding this comment.
Triggers SIGFPE, because H is a uint64 and will overflow to get it to fit within the modulo operation. The cast makes the existing behavior explicit/not-error-y, but I have not checked to see what that means for the hash table generally
There was a problem hiding this comment.
-Can you move this to a seperate
-use C style casts.
-I count 3 instances of the bug, it seems you only have two of them. Can you add the last one. (one in each of Get/Set/Remove)
2 participants
More details in code review comments. These changes were necessary to get microprofile working using GCC on linux, with Vulkan timers enabled.