[AMDGPU] Use directive for kernarg preload header padding (#86004)

Author: kerbowa

llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp

@@ -284,6 +284,15 @@ bool AMDGPUTargetAsmStreamer::EmitHSAMetadata(
   return true;
 }
 
+bool AMDGPUTargetAsmStreamer::EmitKernargPreloadHeader(
+    const MCSubtargetInfo &STI, bool TrapEnabled) {
+  OS << (TrapEnabled ? "\ts_trap 2" : "\ts_endpgm")
+     << " ; Kernarg preload header. Trap with incompatible firmware that "
+        "doesn't support preloading kernel arguments.\n";
+  OS << "\t.fill 63, 4, 0xbf800000 ; s_nop 0\n";
+  return true;
+}
+
 bool AMDGPUTargetAsmStreamer::EmitCodeEnd(const MCSubtargetInfo &STI) {
   const uint32_t Encoded_s_code_end = 0xbf9f0000;
   const uint32_t Encoded_s_nop = 0xbf800000;
@@ -861,18 +870,6 @@ bool AMDGPUTargetELFStreamer::EmitHSAMetadata(msgpack::Document &HSAMetadataDoc,
   return true;
 }
 
-bool AMDGPUTargetAsmStreamer::EmitKernargPreloadHeader(
-    const MCSubtargetInfo &STI, bool TrapEnabled) {
-  const char *TrapInstr = TrapEnabled ? "\ts_trap 2" : "\ts_endpgm";
-  OS << TrapInstr
-     << " ; Trap with incompatible firmware that doesn't "
-        "support preloading kernel arguments.\n";
-  for (int i = 0; i < 63; ++i) {
-    OS << "\ts_nop 0\n";
-  }
-  return true;
-}
-
 bool AMDGPUTargetELFStreamer::EmitKernargPreloadHeader(
     const MCSubtargetInfo &STI, bool TrapEnabled) {
   const uint32_t Encoded_s_nop = 0xbf800000;

llvm/test/CodeGen/AMDGPU/preload-kernarg-header.ll

@@ -1,17 +1,20 @@
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 -amdgpu-kernarg-preload-count=1 -asm-verbose=0 < %s | FileCheck -check-prefixes=GCN,HSA %s
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 -amdgpu-kernarg-preload-count=1 -filetype=obj < %s | llvm-objdump --arch=amdgcn --mcpu=gfx940 --disassemble - | FileCheck -check-prefixes=GCN,HSA %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx940 -amdgpu-kernarg-preload-count=1 -filetype=obj < %s | llvm-objdump --arch=amdgcn --mcpu=gfx940 --disassemble - | FileCheck -check-prefixes=GCN,NON-HSA %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 -amdgpu-kernarg-preload-count=1 -asm-verbose=0 < %s | FileCheck -check-prefixes=GCN,HSA,ASM %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 -amdgpu-kernarg-preload-count=1 -filetype=obj < %s | llvm-objdump --arch=amdgcn --mcpu=gfx940 --disassemble - | FileCheck -check-prefixes=GCN,HSA,OBJ %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx940 -amdgpu-kernarg-preload-count=1 -filetype=obj < %s | llvm-objdump --arch=amdgcn --mcpu=gfx940 --disassemble - | FileCheck -check-prefixes=GCN,NON-HSA,OBJ %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 -amdgpu-kernarg-preload-count=1 -asm-verbose=0 < %s | llvm-mc -triple amdgcn-amd-amdhsa -mcpu=gfx940 -filetype=obj | llvm-objdump --arch=amdgcn --mcpu=gfx940 --disassemble - | FileCheck -check-prefixes=GCN,HSA,OBJ %s
 
 ; GCN: preload_kernarg_header
 ; HSA: s_trap 2
 ; NON-HSA: s_endpgm
-; GCN-COUNT-63: s_nop 0
+; ASM: .fill 63, 4, 0xbf800000 ; s_nop 0
+; OBJ-COUNT-63: s_nop 0
 define amdgpu_kernel void @preload_kernarg_header(ptr %arg) {
     store ptr %arg, ptr %arg
     ret void
 }
 
 ; GCN: non_kernel_function
+; GCN-NOT: s_trap 2
 ; GCN-NOT: s_nop 0
 ; GCN: flat_store
 define void @non_kernel_function(ptr %arg) {