[PATCH] D15492: AMDGPU/SI: Set the code objects private segment size when targeting HSA.

[Tom Stellard via llvm-commits]

tstellarAMD created this revision.
tstellarAMD added a reviewer: arsenm.
tstellarAMD added a subscriber: llvm-commits.
Herald added a subscriber: arsenm.

I'm not sure how things worked before without this.

http://reviews.llvm.org/D15492

Files:
  lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
  test/CodeGen/AMDGPU/private-memory.ll

Index: test/CodeGen/AMDGPU/private-memory.ll
===================================================================
--- test/CodeGen/AMDGPU/private-memory.ll
+++ test/CodeGen/AMDGPU/private-memory.ll
@@ -1,6 +1,7 @@
 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC
 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC
 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
+; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -mtriple=amdgcn-amdhsa -mcpu=kaveri < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-ALLOCA
 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC
 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
 
@@ -18,6 +19,12 @@
 ; SI-PROMOTE: ds_read_b32
 ; SI-PROMOTE: ds_read_b32
 
+; HSA-ALLOCA: .amd_kernel_code_t
+; FIXME: Creating the emergency stack slots causes us to over-estimate scratch
+; by 4 bytes.
+; HSA-ALLOCA: workitem_private_segment_byte_size = 24
+; HSA-ALLOCA: .end_amd_kernel_code_t
+
 ; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0
 ; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0
 define void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
Index: lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -619,6 +619,7 @@
   header.kernarg_segment_byte_size = MFI->ABIArgOffset;
   header.wavefront_sgpr_count = KernelInfo.NumSGPR;
   header.workitem_vgpr_count = KernelInfo.NumVGPR;
+  header.workitem_private_segment_byte_size = KernelInfo.ScratchSize;
 
   AMDGPUTargetStreamer *TS =
       static_cast<AMDGPUTargetStreamer *>(OutStreamer->getTargetStreamer());


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D15492.42715.patch
Type: text/x-patch
Size: 2473 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151214/3ba56462/attachment.bin>