AMDGPU/GFX10: Don't generate s_code_end padding in the asm-printer

Summary:
The purpose of the padding is to guard against stale code being
fetched into the instruction cache by the lowest level prefetching.
We're generating relocatable ELF here, and so the padding should
arguably be added by the linker. This is in fact what Mesa does.

This also fixes multi-part shaders for Mesa.

Change-Id: I6bfede58f20e9f337762ccf39ef9e0e263e69e82

Reviewers: arsenm, rampitec, t-tye

Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D63427

llvm-svn: 363602

2 files changed, 55 insertions, 47 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
index 03c6a671338..5578251bcfc 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -309,7 +309,13 @@ void AMDGPUAsmPrinter::EmitGlobalVariable(const GlobalVariable *GV) {
 bool AMDGPUAsmPrinter::doFinalization(Module &M) {
   CallGraphResourceInfo.clear();
 
-  if (AMDGPU::isGFX10(*getGlobalSTI())) {
+  // Pad with s_code_end to help tools and guard against instruction prefetch
+  // causing stale data in caches. Arguably this should be done by the linker,
+  // which is why this isn't done for Mesa.
+  const MCSubtargetInfo &STI = *getGlobalSTI();
+  if (AMDGPU::isGFX10(STI) &&
+      (STI.getTargetTriple().getOS() == Triple::AMDHSA ||
+       STI.getTargetTriple().getOS() == Triple::AMDPAL)) {
     OutStreamer->SwitchSection(getObjFileLowering().getTextSection());
     getTargetStreamer()->EmitCodeEnd();
   }
diff --git a/llvm/test/CodeGen/AMDGPU/s_code_end.ll b/llvm/test/CodeGen/AMDGPU/s_code_end.ll
index 80f566ba926..2e87098a42f 100644
--- a/llvm/test/CodeGen/AMDGPU/s_code_end.ll
+++ b/llvm/test/CodeGen/AMDGPU/s_code_end.ll
@@ -1,11 +1,13 @@
-; RUN: llc -march=amdgcn -mcpu=gfx1010 -asm-verbose=0 < %s | FileCheck -check-prefixes=GCN,GCN-ASM,GFX10,GFX10-ASM %s
-; RUN: llc -march=amdgcn -mcpu=gfx1010 -filetype=obj < %s | llvm-objdump -arch=amdgcn -mcpu=gfx1010 -disassemble - | FileCheck -check-prefixes=GCN,GCN-OBJ,GFX10,GFX10-OBJ %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -asm-verbose=0 < %s | FileCheck -check-prefixes=GCN,GCN-ASM,GFX10END,GFX10END-ASM %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -filetype=obj < %s | llvm-objdump -arch=amdgcn -mcpu=gfx1010 -disassemble - | FileCheck -check-prefixes=GCN,GCN-OBJ,GFX10END,GFX10END-OBJ %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 -asm-verbose=0 < %s | FileCheck -check-prefixes=GCN,GCN-ASM,GFX10END,GFX10END-ASM %s
+; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -asm-verbose=0 < %s | FileCheck -check-prefixes=GCN,GCN-ASM,GFX10NOEND,GFX10NOEND-ASM %s
+; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1010 -filetype=obj < %s | llvm-objdump -arch=amdgcn -mcpu=gfx1010 -disassemble - | FileCheck -check-prefixes=GCN,GCN-OBJ,GFX10NOEND,GFX10NOEND-OBJ %s
 
 ; GCN:            a_kernel1:
-; GCN-NEXT:               s_endpgm
-; GCN-ASM-NEXT:   [[END_LABEL1:\.Lfunc_end.*]]:
+; GCN:                    s_endpgm
+; GCN-ASM:        [[END_LABEL1:\.Lfunc_end.*]]:
 ; GCN-ASM-NEXT:           .size   a_kernel1, [[END_LABEL1]]-a_kernel1
-; GCN-ASM:                .section        .AMDGPU.config
 
 ; GCN-OBJ-NEXT:           s_nop 0
 
@@ -14,19 +16,17 @@ define amdgpu_kernel void @a_kernel1() {
 }
 
 ; GCN:            a_kernel2:
-; GCN-NEXT:               s_endpgm
-; GCN-ASM-NEXT:   [[END_LABEL2:\.Lfunc_end.*]]:
+; GCN:                    s_endpgm
+; GCN-ASM:        [[END_LABEL2:\.Lfunc_end.*]]:
 ; GCN-ASM-NEXT:           .size   a_kernel2, [[END_LABEL2]]-a_kernel2
-; GCN-ASM:                .section        .AMDGPU.config
 
-; GCN-OBJ-NEXT:   {{^$}}
+; GCN-OBJ:   {{^$}}
 
 define amdgpu_kernel void @a_kernel2() {
   ret void
 }
 
-; GCN-ASM:                .text
-; GCN-ASM-NEXT:           .globl  a_function
+; GCN-ASM:                .globl  a_function
 ; GCN-ASM-NEXT:           .p2align        2
 ; GCN-ASM-NEXT:           .type   a_function,@function
 
@@ -34,46 +34,48 @@ define amdgpu_kernel void @a_kernel2() {
 ; GCN:                    s_setpc_b64
 ; GCN-ASM-NEXT:   [[END_LABEL3:\.Lfunc_end.*]]:
 ; GCN-ASM-NEXT:           .size   a_function, [[END_LABEL3]]-a_function
-; GFX10-ASM:              .p2alignl 6, 3214868480
-; GFX10-ASM-NEXT:         .fill 32, 4, 3214868480
+; GFX10END-ASM:           .p2alignl 6, 3214868480
+; GFX10END-ASM-NEXT:      .fill 32, 4, 3214868480
+; GFX10NOEND-NOT:         .fill
 
-; GFX10-OBJ-NEXT:         s_code_end
+; GFX10NOEND-OBJ-NOT:     s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
 
-; GFX10-OBJ:              s_code_end // 000000000140:
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
+; GFX10END-OBJ:           s_code_end // 000000000140:
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
 
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
 
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
 
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
-; GFX10-OBJ-NEXT:         s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
+; GFX10END-OBJ-NEXT:      s_code_end
 
 define void @a_function() {
   ret void