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

SWDEV-405575: Added gfx941 and gfx942

Browse Source 
Change-Id: I45a49cd64a76d3ae32c209497c70fe27b5be212b


[ROCm/rocprofiler commit: e1285e3fd4]
This commit is contained in:
Giovanni LB
2023-06-13 15:26:06 -03:00
committed by Giovanni Baraldi
parent 62d3c0018c
commit a7e8182a21
11 changed files with 52 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocprofiler/test/tool/gfx_metrics.xml
+4
View File
@@ -631,6 +631,9 @@
<metric name="TCC_EA0_WRREQ_DRAM" block=TCC event=103 descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC)."></metric>
</gfx940>
<gfx941 base="gfx940"></gfx941>
<gfx942 base="gfx940"></gfx942>
<gfx10>
<metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
<metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
@@ -739,3 +742,4 @@
<gfx1101 base="gfx11">
</gfx1101>
projects/rocprofiler/test/tool/metrics.xml
+2
View File
@@ -426,6 +426,8 @@
<gfx90a base="gfx90a_expr"></gfx90a>
#Mi300
<gfx940 base="gfx940_expr"></gfx940>
<gfx941 base="gfx940_expr"></gfx941>
<gfx942 base="gfx940_expr"></gfx942>
#Navi21
<gfx1030 base="gfx1030_expr"></gfx1030>
projects/rocprofiler/test/tool/tool.cpp
+3 -2
View File
@@ -670,7 +670,7 @@ static const kernel_descriptor_t* GetKernelCode(uint64_t kernel_object) {
}
static uint32_t arch_vgpr_count(const AgentInfo &info, const kernel_descriptor_t &kernel_code) {
if (strcmp(info.name, "gfx90a") == 0 || strcmp(info.name, "gfx940") == 0)
if (strcmp(info.name, "gfx90a") == 0 || strncmp(info.name, "gfx94", 5) == 0)
return (AMD_HSA_BITS_GET(kernel_code.compute_pgm_rsrc3, AMD_COMPUTE_PGM_RSRC_THREE_ACCUM_OFFSET) + 1) * 4;
return (AMD_HSA_BITS_GET(kernel_code.compute_pgm_rsrc1, AMD_COMPUTE_PGM_RSRC_ONE_GRANULATED_WORKITEM_VGPR_COUNT) + 1)
@@ -680,7 +680,7 @@ static uint32_t arch_vgpr_count(const AgentInfo &info, const kernel_descriptor_t
static uint32_t accum_vgpr_count(const AgentInfo &info, const kernel_descriptor_t &kernel_code) {
if (strcmp(info.name, "gfx908") == 0)
return arch_vgpr_count(info, kernel_code);
if (strcmp(info.name, "gfx90a") == 0 || strcmp(info.name, "gfx940") == 0)
if (strcmp(info.name, "gfx90a") == 0 || strncmp(info.name, "gfx94", 5) == 0)
return (AMD_HSA_BITS_GET(kernel_code.compute_pgm_rsrc1,
AMD_COMPUTE_PGM_RSRC_ONE_GRANULATED_WORKITEM_VGPR_COUNT) + 1) * 8
- arch_vgpr_count(info, kernel_code);
@@ -737,6 +737,7 @@ void set_kernel_properties(const rocprofiler_callback_data_t* callback_data,
// Kernel disoatch callback
hsa_status_t dispatch_callback(const rocprofiler_callback_data_t* callback_data, void* user_data,
rocprofiler_group_t* group) {
// TODO: return success, make atomic flag
// Passed tool data
callbacks_data_t* tool_data = reinterpret_cast<callbacks_data_t*>(user_data);
// HSA status