mesa/src/amd/compiler/aco_print_asm.cpp

/*
 * Copyright © 2018 Valve Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 */

#include "aco_ir.h"

#include "util/u_debug.h"

#ifdef LLVM_AVAILABLE
#if defined(_MSC_VER) && defined(restrict)
#undef restrict
#endif
#include "llvm/ac_llvm_util.h"

#include "llvm-c/Disassembler.h"
#include <llvm/ADT/StringRef.h>
#include <llvm/MC/MCDisassembler/MCDisassembler.h>
#endif

#include <array>
#include <iomanip>
#include <vector>

namespace aco {
namespace {

std::vector<bool>
get_referenced_blocks(Program* program)
{
   std::vector<bool> referenced_blocks(program->blocks.size());
   referenced_blocks[0] = true;
   for (Block& block : program->blocks) {
      for (unsigned succ : block.linear_succs)
         referenced_blocks[succ] = true;
   }
   return referenced_blocks;
}

void
print_block_markers(FILE* output, Program* program, const std::vector<bool>& referenced_blocks,
                    unsigned* next_block, unsigned pos)
{
   while (*next_block < program->blocks.size() && pos == program->blocks[*next_block].offset) {
      if (referenced_blocks[*next_block])
         fprintf(output, "BB%u:\n", *next_block);
      (*next_block)++;
   }
}

void
print_instr(FILE* output, const std::vector<uint32_t>& binary, char* instr, unsigned size,
            unsigned pos)
{
   fprintf(output, "%-60s ;", instr);

   for (unsigned i = 0; i < size; i++)
      fprintf(output, " %.8x", binary[pos + i]);
   fputc('\n', output);
}

void
print_constant_data(FILE* output, Program* program)
{
   if (program->constant_data.empty())
      return;

   fputs("\n/* constant data */\n", output);
   for (unsigned i = 0; i < program->constant_data.size(); i += 32) {
      fprintf(output, "[%.6u]", i);
      unsigned line_size = std::min<size_t>(program->constant_data.size() - i, 32);
      for (unsigned j = 0; j < line_size; j += 4) {
         unsigned size = std::min<size_t>(program->constant_data.size() - (i + j), 4);
         uint32_t v = 0;
         memcpy(&v, &program->constant_data[i + j], size);
         fprintf(output, " %.8x", v);
      }
      fputc('\n', output);
   }
}

/**
 * Determines the GPU type to use for CLRXdisasm
 */
const char*
to_clrx_device_name(amd_gfx_level gfx_level, radeon_family family)
{
   switch (gfx_level) {
   case GFX6:
      switch (family) {
      case CHIP_TAHITI: return "tahiti";
      case CHIP_PITCAIRN: return "pitcairn";
      case CHIP_VERDE: return "capeverde";
      case CHIP_OLAND: return "oland";
      case CHIP_HAINAN: return "hainan";
      default: return nullptr;
      }
   case GFX7:
      switch (family) {
      case CHIP_BONAIRE: return "bonaire";
      case CHIP_KAVERI: return "gfx700";
      case CHIP_HAWAII: return "hawaii";
      default: return nullptr;
      }
   case GFX8:
      switch (family) {
      case CHIP_TONGA: return "tonga";
      case CHIP_ICELAND: return "iceland";
      case CHIP_CARRIZO: return "carrizo";
      case CHIP_FIJI: return "fiji";
      case CHIP_STONEY: return "stoney";
      case CHIP_POLARIS10: return "polaris10";
      case CHIP_POLARIS11: return "polaris11";
      case CHIP_POLARIS12: return "polaris12";
      case CHIP_VEGAM: return "polaris11";
      default: return nullptr;
      }
   case GFX9:
      switch (family) {
      case CHIP_VEGA10: return "vega10";
      case CHIP_VEGA12: return "vega12";
      case CHIP_VEGA20: return "vega20";
      case CHIP_RAVEN: return "raven";
      default: return nullptr;
      }
   case GFX10:
      switch (family) {
      case CHIP_NAVI10: return "gfx1010";
      case CHIP_NAVI12: return "gfx1011";
      default: return nullptr;
      }
   case GFX10_3:
   case GFX11: return nullptr;
   default: unreachable("Invalid chip class!"); return nullptr;
   }
}

bool
get_branch_target(char** output, Program* program, const std::vector<bool>& referenced_blocks,
                  char** line_start)
{
   unsigned pos;
   if (sscanf(*line_start, ".L%d_0", &pos) != 1)
      return false;
   pos /= 4;
   *line_start = strchr(*line_start, '_') + 2;

   for (Block& block : program->blocks) {
      if (referenced_blocks[block.index] && block.offset == pos) {
         *output += sprintf(*output, "BB%u", block.index);
         return true;
      }
   }
   return false;
}

bool
print_asm_clrx(Program* program, std::vector<uint32_t>& binary, unsigned exec_size, FILE* output)
{
#ifdef _WIN32
   return true;
#else
   char path[] = "/tmp/fileXXXXXX";
   char line[2048], command[128];
   FILE* p;
   int fd;

   const char* gpu_type = to_clrx_device_name(program->gfx_level, program->family);

   /* Dump the binary into a temporary file. */
   fd = mkstemp(path);
   if (fd < 0)
      return true;

   for (unsigned i = 0; i < exec_size; i++) {
      if (write(fd, &binary[i], 4) == -1)
         goto fail;
   }

   sprintf(command, "clrxdisasm --gpuType=%s -r %s", gpu_type, path);

   p = popen(command, "r");
   if (p) {
      if (!fgets(line, sizeof(line), p)) {
         fprintf(output, "clrxdisasm not found\n");
         pclose(p);
         goto fail;
      }

      std::vector<bool> referenced_blocks = get_referenced_blocks(program);
      unsigned next_block = 0;

      char prev_instr[2048];
      unsigned prev_pos = 0;
      do {
         char* line_start = line;
         if (strncmp(line_start, "/*", 2))
            continue;

         unsigned pos;
         if (sscanf(line_start, "/*%x*/", &pos) != 1)
            continue;
         pos /= 4u; /* get the dword position */

         while (strncmp(line_start, "*/", 2))
            line_start++;
         line_start += 2;

         while (line_start[0] == ' ')
            line_start++;
         *strchr(line_start, '\n') = 0;

         if (*line_start == 0)
            continue; /* not an instruction, only a comment */

         if (pos != prev_pos) {
            /* Print the previous instruction, now that we know the encoding size. */
            print_instr(output, binary, prev_instr, pos - prev_pos, prev_pos);
            prev_pos = pos;
         }

         print_block_markers(output, program, referenced_blocks, &next_block, pos);

         char* dest = prev_instr;
         *(dest++) = '\t';
         while (*line_start) {
            if (!strncmp(line_start, ".L", 2) &&
                get_branch_target(&dest, program, referenced_blocks, &line_start))
               continue;
            *(dest++) = *(line_start++);
         }
         *(dest++) = 0;
      } while (fgets(line, sizeof(line), p));

      if (prev_pos != exec_size)
         print_instr(output, binary, prev_instr, exec_size - prev_pos, prev_pos);

      pclose(p);

      print_constant_data(output, program);
   }

   return false;

fail:
   close(fd);
   unlink(path);
   return true;
#endif
}

#ifdef LLVM_AVAILABLE
std::pair<bool, size_t>
disasm_instr(amd_gfx_level gfx_level, LLVMDisasmContextRef disasm, uint32_t* binary,
             unsigned exec_size, size_t pos, char* outline, unsigned outline_size)
{
   size_t l =
      LLVMDisasmInstruction(disasm, (uint8_t*)&binary[pos], (exec_size - pos) * sizeof(uint32_t),
                            pos * 4, outline, outline_size);

   if (gfx_level >= GFX10 && l == 8 && ((binary[pos] & 0xffff0000) == 0xd7610000) &&
       ((binary[pos + 1] & 0x1ff) == 0xff)) {
      /* v_writelane with literal uses 3 dwords but llvm consumes only 2 */
      l += 4;
   }

   bool invalid = false;
   size_t size;
   if (!l &&
       ((gfx_level >= GFX9 &&
         (binary[pos] & 0xffff8000) == 0xd1348000) || /* v_add_u32_e64 + clamp */
        (gfx_level >= GFX10 &&
         (binary[pos] & 0xffff8000) == 0xd7038000) || /* v_add_u16_e64 + clamp */
        (gfx_level <= GFX9 &&
         (binary[pos] & 0xffff8000) == 0xd1268000) || /* v_add_u16_e64 + clamp */
        (gfx_level >= GFX10 && (binary[pos] & 0xffff8000) == 0xd76d8000) || /* v_add3_u32 + clamp */
        (gfx_level == GFX9 && (binary[pos] & 0xffff8000) == 0xd1ff8000)) /* v_add3_u32 + clamp */) {
      strcpy(outline, "\tinteger addition + clamp");
      bool has_literal = gfx_level >= GFX10 && (((binary[pos + 1] & 0x1ff) == 0xff) ||
                                                (((binary[pos + 1] >> 9) & 0x1ff) == 0xff));
      size = 2 + has_literal;
   } else if (gfx_level >= GFX10 && l == 4 && ((binary[pos] & 0xfe0001ff) == 0x020000f9)) {
      strcpy(outline, "\tv_cndmask_b32 + sdwa");
      size = 2;
   } else if (!l) {
      strcpy(outline, "(invalid instruction)");
      size = 1;
      invalid = true;
   } else {
      assert(l % 4 == 0);
      size = l / 4;
   }

#if LLVM_VERSION_MAJOR <= 14
   /* See: https://github.com/GPUOpen-Tools/radeon_gpu_profiler/issues/65 and
    * https://github.com/llvm/llvm-project/issues/38652
    */
   if (invalid) {
      /* do nothing */
   } else if (gfx_level == GFX9 && (binary[pos] & 0xfc024000) == 0xc0024000) {
      /* SMEM with IMM=1 and SOE=1: LLVM ignores SOFFSET */
      size_t len = strlen(outline);

      char imm[16] = {0};
      while (outline[--len] != ' ') ;
      strncpy(imm, outline + len + 1, sizeof(imm) - 1);

      snprintf(outline + len, outline_size - len, " s%u offset:%s", binary[pos + 1] >> 25, imm);
   } else if (gfx_level >= GFX10 && (binary[pos] & 0xfc000000) == 0xf4000000 &&
              (binary[pos + 1] & 0xfe000000) != 0xfa000000) {
      /* SMEM non-NULL SOFFSET: LLVM ignores OFFSET */
      uint32_t offset = binary[pos + 1] & 0x1fffff;
      if (offset) {
         size_t len = strlen(outline);
         snprintf(outline + len, outline_size - len, " offset:0x%x", offset);
      }
   }
#endif

   return std::make_pair(invalid, size);
}

bool
print_asm_llvm(Program* program, std::vector<uint32_t>& binary, unsigned exec_size, FILE* output)
{
   std::vector<bool> referenced_blocks = get_referenced_blocks(program);

   std::vector<llvm::SymbolInfoTy> symbols;
   std::vector<std::array<char, 16>> block_names;
   block_names.reserve(program->blocks.size());
   for (Block& block : program->blocks) {
      if (!referenced_blocks[block.index])
         continue;
      std::array<char, 16> name;
      sprintf(name.data(), "BB%u", block.index);
      block_names.push_back(name);
      symbols.emplace_back(block.offset * 4,
                           llvm::StringRef(block_names[block_names.size() - 1].data()), 0);
   }

   const char* features = "";
   if (program->gfx_level >= GFX10 && program->wave_size == 64) {
      features = "+wavefrontsize64";
   }

   LLVMDisasmContextRef disasm =
      LLVMCreateDisasmCPUFeatures("amdgcn-mesa-mesa3d", ac_get_llvm_processor_name(program->family),
                                  features, &symbols, 0, NULL, NULL);

   size_t pos = 0;
   bool invalid = false;
   unsigned next_block = 0;

   unsigned prev_size = 0;
   unsigned prev_pos = 0;
   unsigned repeat_count = 0;
   while (pos < exec_size) {
      bool new_block =
         next_block < program->blocks.size() && pos == program->blocks[next_block].offset;
      if (pos + prev_size <= exec_size && prev_pos != pos && !new_block &&
          memcmp(&binary[prev_pos], &binary[pos], prev_size * 4) == 0) {
         repeat_count++;
         pos += prev_size;
         continue;
      } else {
         if (repeat_count)
            fprintf(output, "\t(then repeated %u times)\n", repeat_count);
         repeat_count = 0;
      }

      print_block_markers(output, program, referenced_blocks, &next_block, pos);

      char outline[1024];
      std::pair<bool, size_t> res = disasm_instr(program->gfx_level, disasm, binary.data(),
                                                 exec_size, pos, outline, sizeof(outline));
      invalid |= res.first;

      print_instr(output, binary, outline, res.second, pos);

      prev_size = res.second;
      prev_pos = pos;
      pos += res.second;
   }
   assert(next_block == program->blocks.size());

   LLVMDisasmDispose(disasm);

   print_constant_data(output, program);

   return invalid;
}
#endif /* LLVM_AVAILABLE */

} /* end namespace */

bool
check_print_asm_support(Program* program)
{
#ifdef LLVM_AVAILABLE
   if (program->gfx_level >= GFX8) {
      /* LLVM disassembler only supports GFX8+ */
      const char* name = ac_get_llvm_processor_name(program->family);
      const char* triple = "amdgcn--";
      LLVMTargetRef target = ac_get_llvm_target(triple);

      LLVMTargetMachineRef tm = LLVMCreateTargetMachine(
         target, triple, name, "", LLVMCodeGenLevelDefault, LLVMRelocDefault, LLVMCodeModelDefault);

      bool supported = ac_is_llvm_processor_supported(tm, name);
      LLVMDisposeTargetMachine(tm);

      if (supported)
         return true;
   }
#endif

#ifndef _WIN32
   /* Check if CLRX disassembler binary is available and can disassemble the program */
   return to_clrx_device_name(program->gfx_level, program->family) &&
          system("clrxdisasm --version") == 0;
#else
   return false;
#endif
}

/* Returns true on failure */
bool
print_asm(Program* program, std::vector<uint32_t>& binary, unsigned exec_size, FILE* output)
{
#ifdef LLVM_AVAILABLE
   if (program->gfx_level >= GFX8) {
      return print_asm_llvm(program, binary, exec_size, output);
   }
#endif

   return print_asm_clrx(program, binary, exec_size, output);
}

} // namespace aco