AMDGPU/AMDHSA: Set COMPUTE_PGM_RSRC2:LDS_SIZE to 0

[kzhuravl]

Summary

• This field is populated by the CP

[arsenm]

lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

633 ↗

(On Diff #97006)

Needs a comment

test/CodeGen/AMDGPU/lds-size.ll

15 ↗

(On Diff #97006)

Why do we even print this if it's always going to be 0? why is it in the kernel_code_t?

[kzhuravl]

Address review feedback.

[kzhuravl]

test/CodeGen/AMDGPU/lds-size.ll

15 ↗

(On Diff #97006)

Why do we even print this if it's always going to be 0?

I agree that we can omit printing it here. We are doing some restructuring in amd_kernel_code_t area. Can we include this topic there?

why is it in the kernel_code_t?

granulated_lds_size is a part of COMPUTE_PGM_RSRC2 register, and we cannot omit it from there.

[kzhuravl]

Ping.

[arsenm]

LGTM

llvm/trunk/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

ProgInfo.ComputePGMRSrc2 =
       S_00B84C_USER_SGPR(MFI->getNumUserSGPRs()) |
       S_00B84C_TRAP_HANDLER(STM.isTrapHandlerEnabled()) |
       S_00B84C_TGID_X_EN(MFI->hasWorkGroupIDX()) |
       S_00B84C_TGID_Y_EN(MFI->hasWorkGroupIDY()) |
       S_00B84C_TGID_Z_EN(MFI->hasWorkGroupIDZ()) |
       S_00B84C_TG_SIZE_EN(MFI->hasWorkGroupInfo()) |
       S_00B84C_TIDIG_COMP_CNT(TIDIGCompCnt) |
       S_00B84C_EXCP_EN_MSB(0) |
-      S_00B84C_LDS_SIZE(ProgInfo.LDSBlocks) |
+      // For AMDHSA, LDS_SIZE must be zero, as it is populated by the CP.
+      S_00B84C_LDS_SIZE(STM.isAmdHsaOS() ? 0 : ProgInfo.LDSBlocks) |
       S_00B84C_EXCP_EN(0);
 }
 
 static unsigned getRsrcReg(CallingConv::ID CallConv) {
   switch (CallConv) {
   default: LLVM_FALLTHROUGH;
   case CallingConv::AMDGPU_CS: return R_00B848_COMPUTE_PGM_RSRC1;
   case CallingConv::AMDGPU_HS: return R_00B428_SPI_SHADER_PGM_RSRC1_HS;

llvm/trunk/test/CodeGen/AMDGPU/lds-size.ll

 ; RUN: llc -march=amdgcn < %s | FileCheck -check-prefix=ALL -check-prefix=GCN %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa < %s | FileCheck -check-prefix=ALL -check-prefix=HSA %s
 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=ALL -check-prefix=EG %s
 
 ; This test makes sure we do not double count global values when they are
 ; used in different basic blocks.
 
 ; GCN: .long 47180
 ; GCN-NEXT: .long 32900
 
 ; EG: .long 166120
 ; EG-NEXT: .long 1
 ; ALL: {{^}}test:
 
+; HSA: granulated_lds_size = 0
+; HSA: workgroup_group_segment_byte_size = 4
+
 ; GCN: ; LDSByteSize: 4 bytes/workgroup (compile time only)
 @lds = internal unnamed_addr addrspace(3) global i32 undef, align 4
 
 define amdgpu_kernel void @test(i32 addrspace(1)* %out, i32 %cond) {
 entry:
   %0 = icmp eq i32 %cond, 0
   br i1 %0, label %if, label %else