• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21  * SOFTWARE.
22  */
23 
24 #include "ac_binary.h"
25 
26 #include "ac_gpu_info.h"
27 #include "util/u_math.h"
28 #include "util/u_memory.h"
29 
30 #ifndef _WIN32
31 #include <gelf.h>
32 #include <libelf.h>
33 #endif
34 #include <sid.h>
35 #include <stdio.h>
36 
37 #define SPILLED_SGPRS 0x4
38 #define SPILLED_VGPRS 0x8
39 
40 /* Parse configuration data in .AMDGPU.config section format. */
ac_parse_shader_binary_config(const char * data,size_t nbytes,unsigned wave_size,const struct radeon_info * info,struct ac_shader_config * conf)41 void ac_parse_shader_binary_config(const char *data, size_t nbytes, unsigned wave_size,
42                                    const struct radeon_info *info, struct ac_shader_config *conf)
43 {
44    for (size_t i = 0; i < nbytes; i += 8) {
45       unsigned reg = util_le32_to_cpu(*(uint32_t *)(data + i));
46       unsigned value = util_le32_to_cpu(*(uint32_t *)(data + i + 4));
47       switch (reg) {
48       case R_00B028_SPI_SHADER_PGM_RSRC1_PS:
49       case R_00B128_SPI_SHADER_PGM_RSRC1_VS:
50       case R_00B228_SPI_SHADER_PGM_RSRC1_GS:
51       case R_00B848_COMPUTE_PGM_RSRC1:
52       case R_00B428_SPI_SHADER_PGM_RSRC1_HS:
53          if (wave_size == 32 || info->wave64_vgpr_alloc_granularity == 8)
54             conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 8);
55          else
56             conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 4);
57 
58          conf->num_sgprs = MAX2(conf->num_sgprs, (G_00B028_SGPRS(value) + 1) * 8);
59          /* TODO: LLVM doesn't set FLOAT_MODE for non-compute shaders */
60          conf->float_mode = G_00B028_FLOAT_MODE(value);
61          conf->rsrc1 = value;
62          break;
63       case R_00B02C_SPI_SHADER_PGM_RSRC2_PS:
64          conf->lds_size = MAX2(conf->lds_size, G_00B02C_EXTRA_LDS_SIZE(value));
65          /* TODO: LLVM doesn't set SHARED_VGPR_CNT for all shader types */
66          conf->num_shared_vgprs = G_00B02C_SHARED_VGPR_CNT(value);
67          conf->rsrc2 = value;
68          break;
69       case R_00B12C_SPI_SHADER_PGM_RSRC2_VS:
70          conf->num_shared_vgprs = G_00B12C_SHARED_VGPR_CNT(value);
71          conf->rsrc2 = value;
72          break;
73       case R_00B22C_SPI_SHADER_PGM_RSRC2_GS:
74          conf->num_shared_vgprs = G_00B22C_SHARED_VGPR_CNT(value);
75          conf->rsrc2 = value;
76          break;
77       case R_00B42C_SPI_SHADER_PGM_RSRC2_HS:
78          conf->num_shared_vgprs = G_00B42C_SHARED_VGPR_CNT(value);
79          conf->rsrc2 = value;
80          break;
81       case R_00B84C_COMPUTE_PGM_RSRC2:
82          conf->lds_size = MAX2(conf->lds_size, G_00B84C_LDS_SIZE(value));
83          conf->rsrc2 = value;
84          break;
85       case R_00B8A0_COMPUTE_PGM_RSRC3:
86          conf->num_shared_vgprs = G_00B8A0_SHARED_VGPR_CNT(value);
87          conf->rsrc3 = value;
88          break;
89       case R_0286CC_SPI_PS_INPUT_ENA:
90          conf->spi_ps_input_ena = value;
91          break;
92       case R_0286D0_SPI_PS_INPUT_ADDR:
93          conf->spi_ps_input_addr = value;
94          break;
95       case R_0286E8_SPI_TMPRING_SIZE:
96       case R_00B860_COMPUTE_TMPRING_SIZE:
97          if (info->gfx_level >= GFX11)
98             conf->scratch_bytes_per_wave = G_00B860_WAVESIZE(value) * 256;
99          else
100             conf->scratch_bytes_per_wave = G_00B860_WAVESIZE(value) * 1024;
101          break;
102       case SPILLED_SGPRS:
103          conf->spilled_sgprs = value;
104          break;
105       case SPILLED_VGPRS:
106          conf->spilled_vgprs = value;
107          break;
108       default: {
109          static bool printed;
110 
111          if (!printed) {
112             fprintf(stderr,
113                     "Warning: LLVM emitted unknown "
114                     "config register: 0x%x\n",
115                     reg);
116             printed = true;
117          }
118       } break;
119       }
120    }
121 
122    if (!conf->spi_ps_input_addr)
123       conf->spi_ps_input_addr = conf->spi_ps_input_ena;
124 
125    /* GFX 10.3 internally:
126     * - aligns VGPRS to 16 for Wave32 and 8 for Wave64
127     * - aligns LDS to 1024
128     *
129     * For shader-db stats, set num_vgprs that the hw actually uses.
130     */
131    if (info->gfx_level == GFX10_3) {
132       conf->num_vgprs = align(conf->num_vgprs, wave_size == 32 ? 16 : 8);
133    }
134 
135    /* Enable 64-bit and 16-bit denormals, because there is no performance
136     * cost.
137     *
138     * Don't enable denormals for 32-bit floats, because:
139     * - denormals disable output modifiers
140     * - denormals break v_mad_f32
141     * - GFX6 & GFX7 would be very slow
142     */
143    conf->float_mode &= ~V_00B028_FP_32_DENORMS;
144    conf->float_mode |= V_00B028_FP_16_64_DENORMS;
145 }
146