AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[gfx12] Finalize support for gfx12 new counter blocks (#81)

Browse Source 
1. Move all new gfx12 block enums to aql_profile_v2.h. hsa_ven_amd_aqlprofile.h will be left untouched.
2. Re-org counter info in gfx12_block_info.h to be purely alphabetic-ordered for easy comparison between different IP versions. Also use auto-gen block name from IP header files whenever possible to reduce manual editing.
3. Remove unused counter info from graphics blocks.
4. Added UTCL2 and VML2 support
5. Added all gfx12 blocks to ctrl test
This commit is contained in:
Ma, Bing
2025-05-29 15:14:09 -07:00
committed by GitHub
parent 1ed169e30c
commit b20803e95d
15 changed files with 322 additions and 329 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/core/gfx90a_factory.cpp
+3 -3
View File
@@ -37,14 +37,14 @@ class Mi200Factory : public Gfx9Factory {
virtual int GetAccumHiID() const override { return 185; };
protected:
static const GpuBlockInfo* block_table_[HSA_VEN_AMD_AQLPROFILE_BLOCKS_NUMBER];
static const GpuBlockInfo* block_table_[AQLPROFILE_BLOCKS_NUMBER];
};
const GpuBlockInfo* Mi200Factory::block_table_[HSA_VEN_AMD_AQLPROFILE_BLOCKS_NUMBER] = {};
const GpuBlockInfo* Mi200Factory::block_table_[AQLPROFILE_BLOCKS_NUMBER] = {};
Mi200Factory::Mi200Factory(const AgentInfo* agent_info)
: Gfx9Factory(block_table_, sizeof(block_table_), agent_info) {
for (unsigned i = 0; i < HSA_VEN_AMD_AQLPROFILE_BLOCKS_NUMBER; ++i) {
for (unsigned i = 0; i < AQLPROFILE_BLOCKS_NUMBER; ++i) {
const GpuBlockInfo* base_table_ptr = Gfx9Factory::block_table_[i];
if (base_table_ptr == NULL) continue;
GpuBlockInfo* block_info = nullptr;