llvm/lib/Target/AMDGPU/SIProgramInfo.h

   1 //===--- SIProgramInfo.h ----------------------------------------*- C++ -*-===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 /// \file
  10 /// Defines struct to track resource usage and hardware flags for kernels and
  11 /// entry functions.
  12 ///
  13 //
  14 //===----------------------------------------------------------------------===//
  15
  16 #ifndef LLVM_LIB_TARGET_AMDGPU_SIPROGRAMINFO_H
  17 #define LLVM_LIB_TARGET_AMDGPU_SIPROGRAMINFO_H
  18
  19 #include "llvm/IR/CallingConv.h"
  20 #include <cstdint>
  21
  22 namespace llvm {
  23
  24 /// Track resource usage for kernels / entry functions.
  25 struct SIProgramInfo {
  26     // Fields set in PGM_RSRC1 pm4 packet.
  27     uint32_t VGPRBlocks = 0;
  28     uint32_t SGPRBlocks = 0;
  29     uint32_t Priority = 0;
  30     uint32_t FloatMode = 0;
  31     uint32_t Priv = 0;
  32     uint32_t DX10Clamp = 0;
  33     uint32_t DebugMode = 0;
  34     uint32_t IEEEMode = 0;
  35     uint32_t WgpMode = 0; // GFX10+
  36     uint32_t MemOrdered = 0; // GFX10+
  37     uint64_t ScratchSize = 0;
  38
  39     // Fields set in PGM_RSRC2 pm4 packet.
  40     uint32_t LDSBlocks = 0;
  41     uint32_t ScratchBlocks = 0;
  42
  43     uint64_t ComputePGMRSrc2 = 0;
  44     uint64_t ComputePGMRSrc3GFX90A = 0;
  45
  46     uint32_t NumVGPR = 0;
  47     uint32_t NumArchVGPR = 0;
  48     uint32_t NumAccVGPR = 0;
  49     uint32_t AccumOffset = 0;
  50     uint32_t TgSplit = 0;
  51     uint32_t NumSGPR = 0;
  52     uint32_t LDSSize = 0;
  53     bool FlatUsed = false;
  54
  55     // Number of SGPRs that meets number of waves per execution unit request.
  56     uint32_t NumSGPRsForWavesPerEU = 0;
  57
  58     // Number of VGPRs that meets number of waves per execution unit request.
  59     uint32_t NumVGPRsForWavesPerEU = 0;
  60
  61     // Final occupancy.
  62     uint32_t Occupancy = 0;
  63
  64     // Whether there is recursion, dynamic allocas, indirect calls or some other
  65     // reason there may be statically unknown stack usage.
  66     bool DynamicCallStack = false;
  67
  68     // Bonus information for debugging.
  69     bool VCCUsed = false;
  70
  71     SIProgramInfo() = default;
  72
  73     /// Compute the value of the ComputePGMRsrc1 register.
  74     uint64_t getComputePGMRSrc1() const;
  75     uint64_t getPGMRSrc1(CallingConv::ID CC) const;
  76 };
  77
  78 } // namespace llvm
  79
  80 #endif // LLVM_LIB_TARGET_AMDGPU_SIPROGRAMINFO_H