R600: Emit config values in register / value pairs

Reviewed-by: Vincent Lejeune <vljn@ovi.com>
Tested-By: Aaron Watry <awatry@gmail.com>
llvm-svn: 181228

3 files changed, 58 insertions, 5 deletions

diff --git a/llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp b/llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp
index c915f508c44..4c35ecf5c9b 100644
--- a/llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp
+++ b/llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp
@@ -22,6 +22,7 @@
 #include "SIDefines.h"
 #include "SIMachineFunctionInfo.h"
 #include "SIRegisterInfo.h"
+#include "R600Defines.h"
 #include "R600MachineFunctionInfo.h"
 #include "R600RegisterInfo.h"
 #include "llvm/MC/MCContext.h"
@@ -78,6 +79,7 @@ void AMDGPUAsmPrinter::EmitProgramInfoR600(MachineFunction &MF) {
   const R600RegisterInfo * RI =
                 static_cast<const R600RegisterInfo*>(TM.getRegisterInfo());
   R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
+  const AMDGPUSubtarget &STM = TM.getSubtarget<AMDGPUSubtarget>();
 
   for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
                                                   BB != BB_E; ++BB) {
@@ -101,9 +103,33 @@ void AMDGPUAsmPrinter::EmitProgramInfoR600(MachineFunction &MF) {
       }
     }
   }
-  OutStreamer.EmitIntValue(MaxGPR + 1, 4);
-  OutStreamer.EmitIntValue(MFI->StackSize, 4);
-  OutStreamer.EmitIntValue(killPixel, 4);
+
+  unsigned RsrcReg;
+  if (STM.device()->getGeneration() >= AMDGPUDeviceInfo::HD5XXX) {
+    // Evergreen / Northern Islands
+    switch (MFI->ShaderType) {
+    default: // Fall through
+    case ShaderType::COMPUTE:  RsrcReg = R_0288D4_SQ_PGM_RESOURCES_LS; break;
+    case ShaderType::GEOMETRY: RsrcReg = R_028878_SQ_PGM_RESOURCES_GS; break;
+    case ShaderType::PIXEL:    RsrcReg = R_028844_SQ_PGM_RESOURCES_PS; break;
+    case ShaderType::VERTEX:   RsrcReg = R_028860_SQ_PGM_RESOURCES_VS; break;
+    }
+  } else {
+    // R600 / R700
+    switch (MFI->ShaderType) {
+    default: // Fall through
+    case ShaderType::GEOMETRY: // Fall through
+    case ShaderType::COMPUTE:  // Fall through
+    case ShaderType::VERTEX:   RsrcReg = R_028868_SQ_PGM_RESOURCES_VS; break;
+    case ShaderType::PIXEL:    RsrcReg = R_028850_SQ_PGM_RESOURCES_PS; break;
+    }
+  }
+
+  OutStreamer.EmitIntValue(RsrcReg, 4);
+  OutStreamer.EmitIntValue(S_NUM_GPRS(MaxGPR + 1) |
+                           S_STACK_SIZE(MFI->StackSize), 4);
+  OutStreamer.EmitIntValue(R_02880C_DB_SHADER_CONTROL, 4);
+  OutStreamer.EmitIntValue(S_02880C_KILL_ENABLE(killPixel), 4);
 }
 
 void AMDGPUAsmPrinter::EmitProgramInfoSI(MachineFunction &MF) {
diff --git a/llvm/lib/Target/R600/R600Defines.h b/llvm/lib/Target/R600/R600Defines.h
index 303ca739ec6..4041550f8aa 100644
--- a/llvm/lib/Target/R600/R600Defines.h
+++ b/llvm/lib/Target/R600/R600Defines.h
@@ -97,4 +97,30 @@ namespace R600Operands {
 
 }
 
+//===----------------------------------------------------------------------===//
+// Config register definitions
+//===----------------------------------------------------------------------===//
+
+#define R_02880C_DB_SHADER_CONTROL                    0x02880C
+#define   S_02880C_KILL_ENABLE(x)                      (((x) & 0x1) << 6)
+
+// These fields are the same for all shader types and families.
+#define   S_NUM_GPRS(x)                         (((x) & 0xFF) << 0)
+#define   S_STACK_SIZE(x)                       (((x) & 0xFF) << 8)
+//===----------------------------------------------------------------------===//
+// R600, R700 Registers
+//===----------------------------------------------------------------------===//
+
+#define R_028850_SQ_PGM_RESOURCES_PS                 0x028850
+#define R_028868_SQ_PGM_RESOURCES_VS                 0x028868
+
+//===----------------------------------------------------------------------===//
+// Evergreen, Northern Islands Registers
+//===----------------------------------------------------------------------===//
+
+#define R_028844_SQ_PGM_RESOURCES_PS                 0x028844
+#define R_028860_SQ_PGM_RESOURCES_VS                 0x028860
+#define R_028878_SQ_PGM_RESOURCES_GS                 0x028878
+#define R_0288D4_SQ_PGM_RESOURCES_LS                 0x0288d4
+
 #endif // R600DEFINES_H_
diff --git a/llvm/test/CodeGen/R600/elf.r600.ll b/llvm/test/CodeGen/R600/elf.r600.ll
index 31ba84a0b1c..0590efb0915 100644
--- a/llvm/test/CodeGen/R600/elf.r600.ll
+++ b/llvm/test/CodeGen/R600/elf.r600.ll
@@ -5,8 +5,9 @@
 ; ELF-CHECK: Name: .AMDGPU.config
 
 ; CONFIG-CHECK: .section .AMDGPU.config
-; CONFIG-CHECK-NEXT: .long   2
-; CONFIG-CHECK-NEXT: .long   1
+; CONFIG-CHECK-NEXT: .long   166100
+; CONFIG-CHECK-NEXT: .long   258
+; CONFIG-CHECK-NEXT: .long   165900
 ; CONFIG-CHECK-NEXT: .long   0
 define void @test(float addrspace(1)* %out, i32 %p) {
    %i = add i32 %p, 2