[AMDGPU] Set rsrc1 flags for graphics shaders

[Flakebi]

Summary

Before they were only set for compute kernels and compute shaders but
not for other shaders.

[nhaehnle]

llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

1238–1239

All shader types have MEM_ORDERED, and GS and HS also have WGP_MODE.

Annoyingly, the bits aren't always in the same place.

What I would suggest is:

• Remove SIProgramInfo::ComputePGMRsrc1 because it's fully redundant
• Add a getPgmRsrc1 helper function which computes the correct register value based on the calling convention

[Flakebi]

Also set MEM_ORDERED and WGP_MODE for supported PGMRSrc1 registers.

Add helper function to get PGMRSrc1 from SIProgramInfo for all calling conventions.

[Flakebi]

Rebase

[Flakebi]

ping

[arsenm]

llvm/lib/Target/AMDGPU/SIProgramInfo.cpp

40–48

switch?

[sebastian-ne]

Switch from if to switch.

llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

 
 amdhsa::kernel_descriptor_t AMDGPUAsmPrinter::getAmdhsaKernelDescriptor(
     const MachineFunction &MF,
     const SIProgramInfo &PI) const {
   amdhsa::kernel_descriptor_t KernelDescriptor;
   memset(&KernelDescriptor, 0x0, sizeof(KernelDescriptor));
 
   assert(isUInt<32>(PI.ScratchSize));
-  assert(isUInt<32>(PI.ComputePGMRSrc1));
+  assert(isUInt<32>(PI.getComputePGMRSrc1()));
   assert(isUInt<32>(PI.ComputePGMRSrc2));
 
   KernelDescriptor.group_segment_fixed_size = PI.LDSSize;
   KernelDescriptor.private_segment_fixed_size = PI.ScratchSize;
-  KernelDescriptor.compute_pgm_rsrc1 = PI.ComputePGMRSrc1;
+  KernelDescriptor.compute_pgm_rsrc1 = PI.getComputePGMRSrc1();
   KernelDescriptor.compute_pgm_rsrc2 = PI.ComputePGMRSrc2;
   KernelDescriptor.kernel_code_properties = getAmdhsaKernelCodeProperties(MF);
 
   return KernelDescriptor;
 }
 
 bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
   CurrentProgramInfo = SIProgramInfo();
ProgInfo.ScratchBlocks =
               1ULL << ScratchAlignShift) >>
       ScratchAlignShift;
 
   if (getIsaVersion(getGlobalSTI()->getCPU()).Major >= 10) {
     ProgInfo.WgpMode = STM.isCuModeEnabled() ? 0 : 1;
     ProgInfo.MemOrdered = 1;
   }
 
-  ProgInfo.ComputePGMRSrc1 =
-      S_00B848_VGPRS(ProgInfo.VGPRBlocks) |
-      S_00B848_SGPRS(ProgInfo.SGPRBlocks) |
-      S_00B848_PRIORITY(ProgInfo.Priority) |
-      S_00B848_FLOAT_MODE(ProgInfo.FloatMode) |
-      S_00B848_PRIV(ProgInfo.Priv) |
-      S_00B848_DX10_CLAMP(ProgInfo.DX10Clamp) |
-      S_00B848_DEBUG_MODE(ProgInfo.DebugMode) |
-      S_00B848_IEEE_MODE(ProgInfo.IEEEMode) |
-      S_00B848_WGP_MODE(ProgInfo.WgpMode) |
-      S_00B848_MEM_ORDERED(ProgInfo.MemOrdered);
-
   // 0 = X, 1 = XY, 2 = XYZ
   unsigned TIDIGCompCnt = 0;
   if (MFI->hasWorkItemIDZ())
     TIDIGCompCnt = 2;
   else if (MFI->hasWorkItemIDY())
     TIDIGCompCnt = 1;
 
   ProgInfo.ComputePGMRSrc2 =

 void AMDGPUAsmPrinter::EmitProgramInfoSI(const MachineFunction &MF,
                                          const SIProgramInfo &CurrentProgramInfo) {
   const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   unsigned RsrcReg = getRsrcReg(MF.getFunction().getCallingConv());
 
   if (AMDGPU::isCompute(MF.getFunction().getCallingConv())) {
     OutStreamer->emitInt32(R_00B848_COMPUTE_PGM_RSRC1);
 
-    OutStreamer->emitInt32(CurrentProgramInfo.ComputePGMRSrc1);
+    OutStreamer->emitInt32(CurrentProgramInfo.getComputePGMRSrc1());
 
     OutStreamer->emitInt32(R_00B84C_COMPUTE_PGM_RSRC2);
     OutStreamer->emitInt32(CurrentProgramInfo.ComputePGMRSrc2);
 
     OutStreamer->emitInt32(R_00B860_COMPUTE_TMPRING_SIZE);
     OutStreamer->emitInt32(S_00B860_WAVESIZE(CurrentProgramInfo.ScratchBlocks));
 
     // TODO: Should probably note flat usage somewhere. SC emits a "FlatPtr32 =
void AMDGPUAsmPrinter::EmitPALMetadata(const MachineFunction &MF,
        const SIProgramInfo &CurrentProgramInfo) {
   const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   auto CC = MF.getFunction().getCallingConv();
   auto MD = getTargetStreamer()->getPALMetadata();
 
   MD->setEntryPoint(CC, MF.getFunction().getName());
   MD->setNumUsedVgprs(CC, CurrentProgramInfo.NumVGPRsForWavesPerEU);
   MD->setNumUsedSgprs(CC, CurrentProgramInfo.NumSGPRsForWavesPerEU);
-  if (AMDGPU::isCompute(MF.getFunction().getCallingConv())) {
-    MD->setRsrc1(CC, CurrentProgramInfo.ComputePGMRSrc1);
+  MD->setRsrc1(CC, CurrentProgramInfo.getPGMRSrc1(CC));
+  if (AMDGPU::isCompute(CC)) {
     MD->setRsrc2(CC, CurrentProgramInfo.ComputePGMRSrc2);
   } else {
-    MD->setRsrc1(CC, S_00B028_VGPRS(CurrentProgramInfo.VGPRBlocks) |
-        S_00B028_SGPRS(CurrentProgramInfo.SGPRBlocks));
     if (CurrentProgramInfo.ScratchBlocks > 0)
       MD->setRsrc2(CC, S_00B84C_SCRATCH_EN(1));
   }
   // ScratchSize is in bytes, 16 aligned.
   MD->setScratchSize(CC, alignTo(CurrentProgramInfo.ScratchSize, 16));
   if (MF.getFunction().getCallingConv() == CallingConv::AMDGPU_PS) {
     MD->setRsrc2(CC, S_00B02C_EXTRA_LDS_SIZE(CurrentProgramInfo.LDSBlocks));
     MD->setSpiPsInputEna(MFI->getPSInputEnable());
     MD->setSpiPsInputAddr(MFI->getPSInputAddr());
assert(F.getCallingConv() == CallingConv::AMDGPU_KERNEL ||
          F.getCallingConv() == CallingConv::SPIR_KERNEL);
 
   const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   const GCNSubtarget &STM = MF.getSubtarget<GCNSubtarget>();
 
   AMDGPU::initDefaultAMDKernelCodeT(Out, &STM);
 
   Out.compute_pgm_resource_registers =
-      CurrentProgramInfo.ComputePGMRSrc1 |
+      CurrentProgramInfo.getComputePGMRSrc1() |
       (CurrentProgramInfo.ComputePGMRSrc2 << 32);
   Out.code_properties |= AMD_CODE_PROPERTY_IS_PTR64;
 
   if (CurrentProgramInfo.DynamicCallStack)
     Out.code_properties |= AMD_CODE_PROPERTY_IS_DYNAMIC_CALLSTACK;
 
   AMD_HSA_BITS_SET(Out.code_properties,
                    AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE,

llvm/lib/Target/AMDGPU/CMakeLists.txt

add_llvm_target(AMDGPUCodeGen
   SIMachineFunctionInfo.cpp
   SIMachineScheduler.cpp
   SIMemoryLegalizer.cpp
   SIOptimizeExecMasking.cpp
   SIOptimizeExecMaskingPreRA.cpp
   SIPeepholeSDWA.cpp
   SIPostRABundler.cpp
   SIPreEmitPeephole.cpp
+  SIProgramInfo.cpp
   SIRegisterInfo.cpp
   SIRemoveShortExecBranches.cpp
   SIShrinkInstructions.cpp
   SIWholeQuadMode.cpp
   GCNILPSched.cpp
   GCNRegBankReassign.cpp
   GCNNSAReassign.cpp
   GCNDPPCombine.cpp
   SIModeRegister.cpp
   )
 
 add_subdirectory(AsmParser)
 add_subdirectory(Disassembler)
 add_subdirectory(MCTargetDesc)
 add_subdirectory(TargetInfo)
 add_subdirectory(Utils)

llvm/lib/Target/AMDGPU/SIDefines.h

enum DppFiMode {
   DPP8_FI_0 = 0xE9,
   DPP8_FI_1 = 0xEA,
 };
 
 } // namespace DPP
 } // namespace AMDGPU
 
 #define R_00B028_SPI_SHADER_PGM_RSRC1_PS                                0x00B028
+#define   S_00B028_VGPRS(x)                                           (((x) & 0x3F) << 0)
+#define   S_00B028_SGPRS(x)                                           (((x) & 0x0F) << 6)
+#define   S_00B028_MEM_ORDERED(x)                                     (((x) & 0x1) << 25)
+#define   G_00B028_MEM_ORDERED(x)                                     (((x) >> 25) & 0x1)
+#define   C_00B028_MEM_ORDERED                                        0xFDFFFFFF
+
 #define R_00B02C_SPI_SHADER_PGM_RSRC2_PS                                0x00B02C
 #define   S_00B02C_EXTRA_LDS_SIZE(x)                                  (((x) & 0xFF) << 8)
 #define R_00B128_SPI_SHADER_PGM_RSRC1_VS                                0x00B128
+#define   S_00B128_MEM_ORDERED(x)                                     (((x) & 0x1) << 27)
+#define   G_00B128_MEM_ORDERED(x)                                     (((x) >> 27) & 0x1)
+#define   C_00B128_MEM_ORDERED                                        0xF7FFFFFF
+
 #define R_00B228_SPI_SHADER_PGM_RSRC1_GS                                0x00B228
+#define   S_00B228_WGP_MODE(x)                                        (((x) & 0x1) << 27)
+#define   G_00B228_WGP_MODE(x)                                        (((x) >> 27) & 0x1)
+#define   C_00B228_WGP_MODE                                           0xF7FFFFFF
+#define   S_00B228_MEM_ORDERED(x)                                     (((x) & 0x1) << 25)
+#define   G_00B228_MEM_ORDERED(x)                                     (((x) >> 25) & 0x1)
+#define   C_00B228_MEM_ORDERED                                        0xFDFFFFFF
+
 #define R_00B328_SPI_SHADER_PGM_RSRC1_ES                                0x00B328
 #define R_00B428_SPI_SHADER_PGM_RSRC1_HS                                0x00B428
+#define   S_00B428_WGP_MODE(x)                                        (((x) & 0x1) << 26)
+#define   G_00B428_WGP_MODE(x)                                        (((x) >> 26) & 0x1)
+#define   C_00B428_WGP_MODE                                           0xFBFFFFFF
+#define   S_00B428_MEM_ORDERED(x)                                     (((x) & 0x1) << 24)
+#define   G_00B428_MEM_ORDERED(x)                                     (((x) >> 24) & 0x1)
+#define   C_00B428_MEM_ORDERED                                        0xFEFFFFFF
+
 #define R_00B528_SPI_SHADER_PGM_RSRC1_LS                                0x00B528
-#define R_00B848_COMPUTE_PGM_RSRC1                                      0x00B848
-#define   S_00B028_VGPRS(x)                                           (((x) & 0x3F) << 0)
-#define   S_00B028_SGPRS(x)                                           (((x) & 0x0F) << 6)
 
 #define R_00B84C_COMPUTE_PGM_RSRC2                                      0x00B84C
 #define   S_00B84C_SCRATCH_EN(x)                                      (((x) & 0x1) << 0)
 #define   G_00B84C_SCRATCH_EN(x)                                      (((x) >> 0) & 0x1)
 #define   C_00B84C_SCRATCH_EN                                         0xFFFFFFFE
 #define   S_00B84C_USER_SGPR(x)                                       (((x) & 0x1F) << 1)
 #define   G_00B84C_USER_SGPR(x)                                       (((x) >> 1) & 0x1F)
 #define   C_00B84C_USER_SGPR                                          0xFFFFFFC1

llvm/lib/Target/AMDGPU/SIProgramInfo.h

 //===--- SIProgramInfo.h ----------------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file
-/// Defines struct to track resource usage for kernels and entry functions.
+/// Defines struct to track resource usage and hardware flags for kernels and
+/// entry functions.
 ///
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_LIB_TARGET_AMDGPU_SIPROGRAMINFO_H
 #define LLVM_LIB_TARGET_AMDGPU_SIPROGRAMINFO_H
 
+#include "llvm/IR/CallingConv.h"
+#include <cstdint>
+
 namespace llvm {
 
 /// Track resource usage for kernels / entry functions.
 struct SIProgramInfo {
     // Fields set in PGM_RSRC1 pm4 packet.
     uint32_t VGPRBlocks = 0;
     uint32_t SGPRBlocks = 0;
     uint32_t Priority = 0;
     uint32_t FloatMode = 0;
     uint32_t Priv = 0;
     uint32_t DX10Clamp = 0;
     uint32_t DebugMode = 0;
     uint32_t IEEEMode = 0;
     uint32_t WgpMode = 0; // GFX10+
     uint32_t MemOrdered = 0; // GFX10+
     uint64_t ScratchSize = 0;
 
-    uint64_t ComputePGMRSrc1 = 0;
-
     // Fields set in PGM_RSRC2 pm4 packet.
     uint32_t LDSBlocks = 0;
     uint32_t ScratchBlocks = 0;
 
     uint64_t ComputePGMRSrc2 = 0;
 
     uint32_t NumVGPR = 0;
     uint32_t NumArchVGPR = 0;
struct SIProgramInfo {
     // Whether there is recursion, dynamic allocas, indirect calls or some other
     // reason there may be statically unknown stack usage.
     bool DynamicCallStack = false;
 
     // Bonus information for debugging.
     bool VCCUsed = false;
 
     SIProgramInfo() = default;
+
+    /// Compute the value of the ComputePGMRsrc1 register.
+    uint64_t getComputePGMRSrc1() const;
+    uint64_t getPGMRSrc1(CallingConv::ID CC) const;
 };
 
 } // namespace llvm
 
 #endif // LLVM_LIB_TARGET_AMDGPU_SIPROGRAMINFO_H

llvm/lib/Target/AMDGPU/SIProgramInfo.cpp

+//===-- SIProgramInfo.cpp ----------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+///
+/// The SIProgramInfo tracks resource usage and hardware flags for kernels and
+/// entry functions.
+//
+//===----------------------------------------------------------------------===//
+//
+
+#include "SIProgramInfo.h"
+#include "SIDefines.h"
+#include "Utils/AMDGPUBaseInfo.h"
+
+using namespace llvm;
+
+uint64_t SIProgramInfo::getComputePGMRSrc1() const {
+  return S_00B848_VGPRS(VGPRBlocks) | S_00B848_SGPRS(SGPRBlocks) |
+         S_00B848_PRIORITY(Priority) | S_00B848_FLOAT_MODE(FloatMode) |
+         S_00B848_PRIV(Priv) | S_00B848_DX10_CLAMP(DX10Clamp) |
+         S_00B848_DEBUG_MODE(DebugMode) | S_00B848_IEEE_MODE(IEEEMode) |
+         S_00B848_WGP_MODE(WgpMode) | S_00B848_MEM_ORDERED(MemOrdered);
+}
+
+uint64_t SIProgramInfo::getPGMRSrc1(CallingConv::ID CC) const {
+  if (AMDGPU::isCompute(CC)) {
+    return getComputePGMRSrc1();
+  }
+  uint64_t Reg = S_00B848_VGPRS(VGPRBlocks) | S_00B848_SGPRS(SGPRBlocks) |
+                 S_00B848_PRIORITY(Priority) | S_00B848_FLOAT_MODE(FloatMode) |
+                 S_00B848_PRIV(Priv) | S_00B848_DX10_CLAMP(DX10Clamp) |
+                 S_00B848_DEBUG_MODE(DebugMode) | S_00B848_IEEE_MODE(IEEEMode);
+  switch (CC) {
+  case CallingConv::AMDGPU_PS:
+    Reg |= S_00B028_MEM_ORDERED(MemOrdered);
+    break;
+  case CallingConv::AMDGPU_VS:
+    Reg |= S_00B128_MEM_ORDERED(MemOrdered);
+    break;
+  case CallingConv::AMDGPU_GS:
+    Reg |= S_00B228_WGP_MODE(WgpMode) | S_00B228_MEM_ORDERED(MemOrdered);
+    break;
+  case CallingConv::AMDGPU_HS:
+    Reg |= S_00B428_WGP_MODE(WgpMode) | S_00B428_MEM_ORDERED(MemOrdered);
+    break;
+  default:
+    break;
+  }
+  return Reg;
+}

llvm/test/CodeGen/AMDGPU/amdpal-msgpack-default.ll

 ; RUN: llc -mtriple=amdgcn--amdpal -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -enable-var-scope %s
 
 ; amdpal compute shader: check for 0x2e12 (COMPUTE_PGM_RSRC1) in pal metadata
 ; SI-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0x2f0000{{$}}
 ; VI-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0x2f02c0{{$}}
 ; GFX9-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0x2f0000{{$}}
 define amdgpu_cs half @cs_amdpal(half %arg0) {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal evaluation shader: check for 0x2cca (SPI_SHADER_PGM_RSRC1_ES) in pal metadata
-; SI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0{{$}}
-; VI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2c0{{$}}
-; GFX9-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0{{$}}
+; SI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2f0000{{$}}
+; VI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2f02c0{{$}}
+; GFX9-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2f0000{{$}}
 define amdgpu_es half @es_amdpal(half %arg0) {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal geometry shader: check for 0x2c8a (SPI_SHADER_PGM_RSRC1_GS) in pal metadata
-; SI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0{{$}}
-; VI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2c0{{$}}
-; GFX9-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0{{$}}
+; SI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2f0000{{$}}
+; VI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2f02c0{{$}}
+; GFX9-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2f0000{{$}}
 define amdgpu_gs half @gs_amdpal(half %arg0) {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal hull shader: check for 0x2d0a (SPI_SHADER_PGM_RSRC1_HS) in pal metadata
-; SI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0{{$}}
-; VI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2c0{{$}}
-; GFX9-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0{{$}}
+; SI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2f0000{{$}}
+; VI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2f02c0{{$}}
+; GFX9-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2f0000{{$}}
 define amdgpu_hs half @hs_amdpal(half %arg0) {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal load shader: check for 0x2d4a (SPI_SHADER_PGM_RSRC1_LS) in pal metadata
-; SI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0{{$}}
-; VI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2c0{{$}}
-; GFX9-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0{{$}}
+; SI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2f0000{{$}}
+; VI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2f02c0{{$}}
+; GFX9-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2f0000{{$}}
 define amdgpu_ls half @ls_amdpal(half %arg0) {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal pixel shader: check for 0x2c0a (SPI_SHADER_PGM_RSRC1_PS) in pal metadata
 ; below.
-; SI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0{{$}}
-; VI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2c0{{$}}
-; GFX9-DAG:         0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0{{$}}
+; SI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2f0000{{$}}
+; VI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2f02c0{{$}}
+; GFX9-DAG:         0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2f0000{{$}}
 define amdgpu_ps half @ps_amdpal(half %arg0) {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal vertex shader: check for 45352 (SPI_SHADER_PGM_RSRC1_VS) in pal metadata
-; SI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0{{$}}
-; VI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2c0{{$}}
-; GFX9-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0{{$}}
+; SI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2f0000{{$}}
+; VI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2f02c0{{$}}
+; GFX9-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2f0000{{$}}
 define amdgpu_vs half @vs_amdpal(half %arg0) {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdgpu.pal.metadata.msgpack represents this:
 ;
 ;   .amdgpu_pal_metadata

llvm/test/CodeGen/AMDGPU/amdpal-msgpack-denormal.ll

 ; RUN: llc -mtriple=amdgcn--amdpal -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -enable-var-scope %s
 
 ; amdpal compute shader: check for 0x2e12 (COMPUTE_PGM_RSRC1) in pal metadata
 ; SI-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0x2c0000{{$}}
 ; VI-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0x2c02c0{{$}}
 ; GFX9-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0x2c0000{{$}}
 define amdgpu_cs half @cs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal evaluation shader: check for 0x2cca (SPI_SHADER_PGM_RSRC1_ES) in pal metadata
-; SI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0{{$}}
-; VI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2c0{{$}}
-; GFX9-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0{{$}}
+; SI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2c0000{{$}}
+; VI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2c02c0{{$}}
+; GFX9-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2c0000{{$}}
 define amdgpu_es half @es_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal geometry shader: check for 0x2c8a (SPI_SHADER_PGM_RSRC1_GS) in pal metadata
-; SI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0{{$}}
-; VI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2c0{{$}}
-; GFX9-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0{{$}}
+; SI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2c0000{{$}}
+; VI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2c02c0{{$}}
+; GFX9-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2c0000{{$}}
 define amdgpu_gs half @gs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal hull shader: check for 0x2d0a (SPI_SHADER_PGM_RSRC1_HS) in pal metadata
-; SI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0{{$}}
-; VI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2c0{{$}}
-; GFX9-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0{{$}}
+; SI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2c0000{{$}}
+; VI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2c02c0{{$}}
+; GFX9-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2c0000{{$}}
 define amdgpu_hs half @hs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal load shader: check for 0x2d4a (SPI_SHADER_PGM_RSRC1_LS) in pal metadata
-; SI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0{{$}}
-; VI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2c0{{$}}
-; GFX9-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0{{$}}
+; SI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2c0000{{$}}
+; VI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2c02c0{{$}}
+; GFX9-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2c0000{{$}}
 define amdgpu_ls half @ls_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal pixel shader: check for 0x2c0a (SPI_SHADER_PGM_RSRC1_PS) in pal metadata
 ; below.
-; SI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0{{$}}
-; VI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2c0{{$}}
-; GFX9-DAG:         0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0{{$}}
+; SI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2c0000{{$}}
+; VI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2c02c0{{$}}
+; GFX9-DAG:         0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2c0000{{$}}
 define amdgpu_ps half @ps_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal vertex shader: check for 45352 (SPI_SHADER_PGM_RSRC1_VS) in pal metadata
-; SI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0{{$}}
-; VI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2c0{{$}}
-; GFX9-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0{{$}}
+; SI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2c0000{{$}}
+; VI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2c02c0{{$}}
+; GFX9-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2c0000{{$}}
 define amdgpu_vs half @vs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
 
 ; amdgpu.pal.metadata.msgpack represents this:

llvm/test/CodeGen/AMDGPU/amdpal-msgpack-dx10-clamp.ll

 ; RUN: llc -mtriple=amdgcn--amdpal -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -enable-var-scope %s
 
 ; amdpal compute shader: check for 0x2e12 (COMPUTE_PGM_RSRC1) in pal metadata
 ; SI-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0xf0000{{$}}
 ; VI-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0xf02c0{{$}}
 ; GFX9-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0xf0000{{$}}
 define amdgpu_cs half @cs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal evaluation shader: check for 0x2cca (SPI_SHADER_PGM_RSRC1_ES) in pal metadata
-; SI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0{{$}}
-; VI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2c0{{$}}
-; GFX9-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0{{$}}
+; SI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0xf0000{{$}}
+; VI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0xf02c0{{$}}
+; GFX9-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0xf0000{{$}}
 define amdgpu_es half @es_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal geometry shader: check for 0x2c8a (SPI_SHADER_PGM_RSRC1_GS) in pal metadata
-; SI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0{{$}}
-; VI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2c0{{$}}
-; GFX9-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0{{$}}
+; SI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0xf0000{{$}}
+; VI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0xf02c0{{$}}
+; GFX9-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0xf0000{{$}}
 define amdgpu_gs half @gs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal hull shader: check for 0x2d0a (SPI_SHADER_PGM_RSRC1_HS) in pal metadata
-; SI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0{{$}}
-; VI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2c0{{$}}
-; GFX9-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0{{$}}
+; SI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0xf0000{{$}}
+; VI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0xf02c0{{$}}
+; GFX9-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0xf0000{{$}}
 define amdgpu_hs half @hs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal load shader: check for 0x2d4a (SPI_SHADER_PGM_RSRC1_LS) in pal metadata
-; SI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0{{$}}
-; VI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2c0{{$}}
-; GFX9-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0{{$}}
+; SI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0xf0000{{$}}
+; VI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0xf02c0{{$}}
+; GFX9-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0xf0000{{$}}
 define amdgpu_ls half @ls_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal pixel shader: check for 0x2c0a (SPI_SHADER_PGM_RSRC1_PS) in pal metadata
 ; below.
-; SI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0{{$}}
-; VI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2c0{{$}}
-; GFX9-DAG:         0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0{{$}}
+; SI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0xf0000{{$}}
+; VI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0xf02c0{{$}}
+; GFX9-DAG:         0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0xf0000{{$}}
 define amdgpu_ps half @ps_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal vertex shader: check for 45352 (SPI_SHADER_PGM_RSRC1_VS) in pal metadata
-; SI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0{{$}}
-; VI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2c0{{$}}
-; GFX9-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0{{$}}
+; SI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0xf0000{{$}}
+; VI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0xf02c0{{$}}
+; GFX9-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0xf0000{{$}}
 define amdgpu_vs half @vs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 attributes #0 = { "amdgpu-dx10-clamp"="false" }
 
 ; amdgpu.pal.metadata.msgpack represents this:

llvm/test/CodeGen/AMDGPU/amdpal-msgpack-ieee.ll

 ; RUN: llc -mtriple=amdgcn--amdpal -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -enable-var-scope %s
 
 ; amdpal compute shader: check for 0x2e12 (COMPUTE_PGM_RSRC1) in pal metadata
 ; SI-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0xaf0000{{$}}
 ; VI-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0xaf02c0{{$}}
 ; GFX9-DAG: 0x2e12 (COMPUTE_PGM_RSRC1): 0xaf0000{{$}}
 define amdgpu_cs half @cs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal evaluation shader: check for 0x2cca (SPI_SHADER_PGM_RSRC1_ES) in pal metadata
-; SI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0{{$}}
-; VI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0x2c0{{$}}
-; GFX9-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0{{$}}
+; SI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0xaf0000{{$}}
+; VI-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0xaf02c0{{$}}
+; GFX9-DAG: 0x2cca (SPI_SHADER_PGM_RSRC1_ES): 0xaf0000{{$}}
 define amdgpu_es half @es_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal geometry shader: check for 0x2c8a (SPI_SHADER_PGM_RSRC1_GS) in pal metadata
-; SI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0{{$}}
-; VI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0x2c0{{$}}
-; GFX9-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0{{$}}
+; SI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0xaf0000{{$}}
+; VI-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0xaf02c0{{$}}
+; GFX9-DAG: 0x2c8a (SPI_SHADER_PGM_RSRC1_GS): 0xaf0000{{$}}
 define amdgpu_gs half @gs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal hull shader: check for 0x2d0a (SPI_SHADER_PGM_RSRC1_HS) in pal metadata
-; SI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0{{$}}
-; VI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0x2c0{{$}}
-; GFX9-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0{{$}}
+; SI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0xaf0000{{$}}
+; VI-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0xaf02c0{{$}}
+; GFX9-DAG: 0x2d0a (SPI_SHADER_PGM_RSRC1_HS): 0xaf0000{{$}}
 define amdgpu_hs half @hs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal load shader: check for 0x2d4a (SPI_SHADER_PGM_RSRC1_LS) in pal metadata
-; SI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0{{$}}
-; VI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0x2c0{{$}}
-; GFX9-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0{{$}}
+; SI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0xaf0000{{$}}
+; VI-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0xaf02c0{{$}}
+; GFX9-DAG: 0x2d4a (SPI_SHADER_PGM_RSRC1_LS): 0xaf0000{{$}}
 define amdgpu_ls half @ls_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal pixel shader: check for 0x2c0a (SPI_SHADER_PGM_RSRC1_PS) in pal metadata
 ; below.
-; SI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0{{$}}
-; VI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0x2c0{{$}}
-; GFX9-DAG:         0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0{{$}}
+; SI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0xaf0000{{$}}
+; VI-DAG:           0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0xaf02c0{{$}}
+; GFX9-DAG:         0x2c0a (SPI_SHADER_PGM_RSRC1_PS): 0xaf0000{{$}}
 define amdgpu_ps half @ps_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 ; amdpal vertex shader: check for 45352 (SPI_SHADER_PGM_RSRC1_VS) in pal metadata
-; SI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0{{$}}
-; VI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0x2c0{{$}}
-; GFX9-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0{{$}}
+; SI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0xaf0000{{$}}
+; VI-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0xaf02c0{{$}}
+; GFX9-DAG: 0x2c4a (SPI_SHADER_PGM_RSRC1_VS): 0xaf0000{{$}}
 define amdgpu_vs half @vs_amdpal(half %arg0) #0 {
   %add = fadd half %arg0, 1.0
   ret half %add
 }
 
 attributes #0 = { "amdgpu-ieee"="true" }
 
 ; amdgpu.pal.metadata.msgpack represents this: