src/amd/common/ac_binary.c

/*
 * Copyright 2014 Advanced Micro Devices, Inc.
 *
 * SPDX-License-Identifier: MIT
 */

#include "ac_binary.h"

#include "ac_gpu_info.h"
#include "util/u_math.h"
#include "util/u_memory.h"

#include <sid.h>
#include <stdio.h>

#define SPILLED_SGPRS 0x4
#define SPILLED_VGPRS 0x8

/* Parse configuration data in .AMDGPU.config section format. */
void ac_parse_shader_binary_config(const char *data, size_t nbytes, unsigned wave_size,
                                   const struct radeon_info *info, struct ac_shader_config *conf)
{
   for (size_t i = 0; i < nbytes; i += 8) {
      unsigned reg = util_le32_to_cpu(*(uint32_t *)(data + i));
      unsigned value = util_le32_to_cpu(*(uint32_t *)(data + i + 4));
      switch (reg) {
      case R_00B028_SPI_SHADER_PGM_RSRC1_PS:
      case R_00B128_SPI_SHADER_PGM_RSRC1_VS:
      case R_00B228_SPI_SHADER_PGM_RSRC1_GS:
      case R_00B848_COMPUTE_PGM_RSRC1:
      case R_00B428_SPI_SHADER_PGM_RSRC1_HS:
         if (wave_size == 32 || info->wave64_vgpr_alloc_granularity == 8)
            conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 8);
         else
            conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 4);

         conf->num_sgprs = MAX2(conf->num_sgprs, (G_00B028_SGPRS(value) + 1) * 8);
         /* TODO: LLVM doesn't set FLOAT_MODE for non-compute shaders */
         conf->float_mode = G_00B028_FLOAT_MODE(value);
         conf->rsrc1 = value;
         break;
      case R_00B02C_SPI_SHADER_PGM_RSRC2_PS:
         conf->lds_size = MAX2(conf->lds_size, G_00B02C_EXTRA_LDS_SIZE(value));
         /* TODO: LLVM doesn't set SHARED_VGPR_CNT for all shader types */
         conf->num_shared_vgprs = G_00B02C_SHARED_VGPR_CNT(value);
         conf->rsrc2 = value;
         break;
      case R_00B12C_SPI_SHADER_PGM_RSRC2_VS:
         conf->num_shared_vgprs = G_00B12C_SHARED_VGPR_CNT(value);
         conf->rsrc2 = value;
         break;
      case R_00B22C_SPI_SHADER_PGM_RSRC2_GS:
         conf->num_shared_vgprs = G_00B22C_SHARED_VGPR_CNT(value);
         conf->rsrc2 = value;
         break;
      case R_00B42C_SPI_SHADER_PGM_RSRC2_HS:
         conf->num_shared_vgprs = G_00B42C_SHARED_VGPR_CNT(value);
         conf->rsrc2 = value;
         break;
      case R_00B84C_COMPUTE_PGM_RSRC2:
         conf->lds_size = MAX2(conf->lds_size, G_00B84C_LDS_SIZE(value));
         conf->rsrc2 = value;
         break;
      case R_00B8A0_COMPUTE_PGM_RSRC3:
         conf->num_shared_vgprs = G_00B8A0_SHARED_VGPR_CNT(value);
         conf->rsrc3 = value;
         break;
      case R_02865C_SPI_PS_INPUT_ENA:
      case R_0286CC_SPI_PS_INPUT_ENA:
         conf->spi_ps_input_ena = value;
         break;
      case R_028660_SPI_PS_INPUT_ADDR:
      case R_0286D0_SPI_PS_INPUT_ADDR:
         conf->spi_ps_input_addr = value;
         break;
      case R_0286E8_SPI_TMPRING_SIZE:
      case R_00B860_COMPUTE_TMPRING_SIZE:
         if (info->gfx_level >= GFX11)
            conf->scratch_bytes_per_wave = G_00B860_WAVESIZE(value) * 256;
         else
            conf->scratch_bytes_per_wave = G_00B860_WAVESIZE(value) * 1024;
         break;
      case SPILLED_SGPRS:
         conf->spilled_sgprs = value;
         break;
      case SPILLED_VGPRS:
         conf->spilled_vgprs = value;
         break;
      default: {
         static bool printed;

         if (!printed) {
            fprintf(stderr,
                    "Warning: LLVM emitted unknown "
                    "config register: 0x%x\n",
                    reg);
            printed = true;
         }
      } break;
      }
   }

   if (!conf->spi_ps_input_addr)
      conf->spi_ps_input_addr = conf->spi_ps_input_ena;

   /* Enable 64-bit and 16-bit denormals, because there is no performance
    * cost.
    *
    * Don't enable denormals for 32-bit floats, because:
    * - denormals disable output modifiers
    * - denormals break v_mad_f32
    * - GFX6 & GFX7 would be very slow
    */
   conf->float_mode &= ~V_00B028_FP_32_DENORMS;
   conf->float_mode |= V_00B028_FP_16_64_DENORMS;
}

unsigned ac_align_shader_binary_for_prefetch(const struct radeon_info *info, unsigned size)
{
   /* The SQ fetches up to N cache lines of 16 dwords
    * ahead of the PC, configurable by SH_MEM_CONFIG and
    * S_INST_PREFETCH. This can cause two issues:
    *
    * (1) Crossing a page boundary to an unmapped page. The logic
    *     does not distinguish between a required fetch and a "mere"
    *     prefetch and will fault.
    *
    * (2) Prefetching instructions that will be changed for a
    *     different shader.
    *
    * (2) is not currently an issue because we flush the I$ at IB
    * boundaries, but (1) needs to be addressed. Due to buffer
    * suballocation, we just play it safe.
    */
   unsigned prefetch_distance = 0;

   if (!info->has_graphics && info->family >= CHIP_MI200)
      prefetch_distance = 16;
   else if (info->gfx_level >= GFX10)
      prefetch_distance = 3;

   if (prefetch_distance) {
      if (info->gfx_level >= GFX11)
         size = align(size + prefetch_distance * 64, 128);
      else
         size = align(size + prefetch_distance * 64, 64);
   }

   return size;
}