Skip to content

Commit 1637c07

Browse files
authored
[openmp][amdgpu] Add DynamicLdsSize to AMDGPUImplicitArgsTy (#65325)
#65273 "hidden_dynamic_lds_size" argument will be added in the reserved section at offset 120 of the implicit argument layout Add DynamicLdsSize to AMDGPUImplicitArgsTy struct at offset 120 and fill the dynamic LDS size before kernel launch.
1 parent 2873060 commit 1637c07

File tree

2 files changed

+4
-1
lines changed

2 files changed

+4
-1
lines changed

openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3203,6 +3203,7 @@ Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice,
32033203
ImplArgs->GroupSizeY = 1;
32043204
ImplArgs->GroupSizeZ = 1;
32053205
ImplArgs->GridDims = 1;
3206+
ImplArgs->DynamicLdsSize = KernelArgs.DynCGroupMem;
32063207
}
32073208

32083209
// Push the kernel launch into the stream.

openmp/libomptarget/plugins-nextgen/amdgpu/utils/UtilitiesRTL.h

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -45,7 +45,9 @@ struct AMDGPUImplicitArgsTy {
4545
uint16_t GroupSizeZ;
4646
uint8_t Unused0[46]; // 46 byte offset.
4747
uint16_t GridDims;
48-
uint8_t Unused1[190]; // 190 byte offset.
48+
uint8_t Unused1[54]; // 54 byte offset.
49+
uint32_t DynamicLdsSize;
50+
uint8_t Unused2[132]; // 132 byte offset.
4951
};
5052

5153
// Dummy struct for COV4 implicitargs.

0 commit comments

Comments
 (0)