AMDGPU: Don't emit amd_kernel_code_t for callable functions

This is inserted directly in the text section. The relocation
for the function ends up resolving to the beginning of the
amd_kernel_code_t header rather than the actual function
entry point.

Also skip some of the comments for initialization
that only makes sense for kernels.

llvm-svn: 300736

llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

@@ -144,6 +144,10 @@ bool AMDGPUAsmPrinter::isBlockOnlyReachableByFallthrough(
 }
 
 void AMDGPUAsmPrinter::EmitFunctionBodyStart() {
+  const AMDGPUMachineFunction *MFI = MF->getInfo<AMDGPUMachineFunction>();
+  if (!MFI->isEntryFunction())
+    return;
+
   const AMDGPUSubtarget &STM = MF->getSubtarget<AMDGPUSubtarget>();
   SIProgramInfo KernelInfo;
   amd_kernel_code_t KernelCode;
@@ -222,13 +226,19 @@ bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
     OutStreamer->SwitchSection(CommentSection);
 
     if (STM.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) {
-      OutStreamer->emitRawComment(" Kernel info:", false);
+      if (MFI->isEntryFunction()) {
+        OutStreamer->emitRawComment(" Kernel info:", false);
+      } else {
+        OutStreamer->emitRawComment(" Function info:", false);
+      }
+
       OutStreamer->emitRawComment(" codeLenInByte = " +
                                   Twine(getFunctionCodeSize(MF)), false);
       OutStreamer->emitRawComment(" NumSgprs: " + Twine(KernelInfo.NumSGPR),
                                   false);
       OutStreamer->emitRawComment(" NumVgprs: " + Twine(KernelInfo.NumVGPR),
                                   false);
+
       OutStreamer->emitRawComment(" FloatMode: " + Twine(KernelInfo.FloatMode),
                                   false);
       OutStreamer->emitRawComment(" IeeeMode: " + Twine(KernelInfo.IEEEMode),
@@ -238,6 +248,9 @@ bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
       OutStreamer->emitRawComment(" LDSByteSize: " + Twine(KernelInfo.LDSSize) +
                                   " bytes/workgroup (compile time only)", false);
 
+      if (!MFI->isEntryFunction())
+        return false;
+
       OutStreamer->emitRawComment(" SGPRBlocks: " +
                                   Twine(KernelInfo.SGPRBlocks), false);
       OutStreamer->emitRawComment(" VGPRBlocks: " +

llvm/test/CodeGen/AMDGPU/hsa-func.ll

@@ -27,7 +27,7 @@
 
 ; ELF: Symbol {
 ; ELF: Name: simple
-; ELF: Size: 292
+; ELF: Size: 44
 ; ELF: Type: Function (0x2)
 ; ELF: }
 
@@ -40,11 +40,10 @@
 ; HSA: .globl simple
 ; HSA: .p2align 2
 ; HSA: {{^}}simple:
-; HSA: .amd_kernel_code_t
+; HSA-NOT: amd_kernel_code_t
-; HSA: enable_sgpr_private_segment_buffer = 1
+
-; HSA: enable_sgpr_kernarg_segment_ptr = 1
+; FIXME: Check this isn't a kernarg load when calling convention implemented.
-; HSA: .end_amd_kernel_code_t
+; XHSA-NOT: s_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[4:5], 0x0
-; HSA: s_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[4:5], 0x0
 
 ; Make sure we are setting the ATC bit:
 ; HSA-CI: s_mov_b32 s[[HI:[0-9]]], 0x100f000
@@ -55,7 +54,8 @@
 
 ; HSA: .Lfunc_end0:
 ; HSA: .size   simple, .Lfunc_end0-simple
-
+; HSA: ; Function info:
+; HSA-NOT: COMPUTE_PGM_RSRC2
 define void @simple(i32 addrspace(1)* %out) {
 entry:
   store i32 0, i32 addrspace(1)* %out