2 * Copyright © 2017 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining
5 * a copy of this software and associated documentation files (the
6 * "Software"), to deal in the Software without restriction, including
7 * without limitation the rights to use, copy, modify, merge, publish,
8 * distribute, sub license, and/or sell copies of the Software, and to
9 * permit persons to whom the Software is furnished to do so, subject to
10 * the following conditions:
12 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
13 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
14 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
15 * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
16 * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
18 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19 * USE OR OTHER DEALINGS IN THE SOFTWARE.
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
32 #include "amd_family.h"
38 struct amdgpu_gpu_info
;
41 /* PCI info: domain:bus:dev:func */
49 const char *marketing_name
;
52 enum radeon_family family
;
53 enum chip_class chip_class
;
55 uint32_t chip_external_rev
;
56 bool has_graphics
; /* false if the chip is compute-only */
57 uint32_t num_compute_rings
;
58 uint32_t num_sdma_rings
;
59 uint32_t clock_crystal_freq
;
60 uint32_t tcc_cache_line_size
;
62 bool has_distributed_tess
;
63 bool has_dcc_constant_encode
;
64 bool has_rbplus
; /* if RB+ registers exist */
65 bool rbplus_allowed
; /* if RB+ is allowed */
66 bool has_load_ctx_reg_pkt
;
67 bool has_out_of_order_rast
;
68 bool cpdma_prefetch_writes_memory
;
69 uint32_t pbb_max_alloc_count
;
70 uint32_t num_sdp_interfaces
;
72 /* There are 2 display DCC codepaths, because display expects unaligned DCC. */
73 /* Disable RB and pipe alignment to skip the retile blit. (1 RB chips only) */
74 bool use_display_dcc_unaligned
;
75 /* Allocate both aligned and unaligned DCC and use the retile blit. */
76 bool use_display_dcc_with_retile_blit
;
79 uint32_t pte_fragment_size
;
80 uint32_t gart_page_size
;
83 uint64_t vram_vis_size
;
85 unsigned gds_gfx_partition_size
;
86 uint64_t max_alloc_size
;
87 uint32_t min_alloc_size
;
88 uint32_t address32_hi
;
89 bool has_dedicated_vram
;
90 bool r600_has_virtual_memory
;
93 bool gfx_ib_pad_with_type2
;
94 unsigned ib_start_alignment
;
95 uint32_t me_fw_version
;
96 uint32_t me_fw_feature
;
97 uint32_t pfp_fw_version
;
98 uint32_t pfp_fw_feature
;
99 uint32_t ce_fw_version
;
100 uint32_t ce_fw_feature
;
102 /* Multimedia info. */
104 bool uvd_enc_supported
;
105 uint32_t uvd_fw_version
;
106 uint32_t vce_fw_version
;
107 uint32_t vce_harvest_config
;
109 /* Kernel & winsys capabilities. */
110 uint32_t drm_major
; /* version */
112 uint32_t drm_patchlevel
;
116 bool has_syncobj_wait_for_submit
;
117 bool has_fence_to_handle
;
118 bool has_ctx_priority
;
119 bool has_local_buffers
;
120 bool kernel_flushes_hdp_before_ib
;
121 bool htile_cmask_support_1d_tiling
;
122 bool si_TA_CS_BC_BASE_ADDR_allowed
;
123 bool has_bo_metadata
;
124 bool has_gpu_reset_status_query
;
125 bool has_eqaa_surface_allocator
;
126 bool has_format_bc1_through_bc7
;
127 bool kernel_flushes_tc_l2_after_ib
;
128 bool has_indirect_compute_dispatch
;
129 bool has_unaligned_shader_loads
;
130 bool has_sparse_vm_mappings
;
132 bool has_read_registers_query
;
133 bool has_gds_ordered_append
;
134 bool has_scheduled_fence_dependency
;
137 uint32_t r600_max_quad_pipes
; /* wave size / 16 */
138 uint32_t max_shader_clock
;
139 uint32_t num_good_compute_units
;
140 uint32_t num_good_cu_per_sh
;
141 uint32_t num_tcc_blocks
;
142 uint32_t max_se
; /* shader engines */
143 uint32_t max_sh_per_se
; /* shader arrays per shader engine */
145 /* Render backends (color + depth blocks). */
146 uint32_t r300_num_gb_pipes
;
147 uint32_t r300_num_z_pipes
;
148 uint32_t r600_gb_backend_map
; /* R600 harvest config */
149 bool r600_gb_backend_map_valid
;
150 uint32_t r600_num_banks
;
151 uint32_t gb_addr_config
;
152 uint32_t pa_sc_tile_steering_override
; /* CLEAR_STATE also sets this */
153 uint32_t num_render_backends
;
154 uint32_t num_tile_pipes
; /* pipe count from PIPE_CONFIG */
155 uint32_t pipe_interleave_bytes
;
156 uint32_t enabled_rb_mask
; /* GCN harvest config */
157 uint64_t max_alignment
; /* from addrlib */
160 uint32_t si_tile_mode_array
[32];
161 uint32_t cik_macrotile_mode_array
[16];
164 bool has_gfx9_scissor_bug
;
165 bool has_tc_compat_zrange_bug
;
166 bool has_msaa_sample_loc_bug
;
167 bool has_ls_vgpr_init_bug
;
170 bool ac_query_gpu_info(int fd
, void *dev_p
,
171 struct radeon_info
*info
,
172 struct amdgpu_gpu_info
*amdinfo
);
174 void ac_compute_driver_uuid(char *uuid
, size_t size
);
176 void ac_compute_device_uuid(struct radeon_info
*info
, char *uuid
, size_t size
);
177 void ac_print_gpu_info(struct radeon_info
*info
);
178 int ac_get_gs_table_depth(enum chip_class chip_class
, enum radeon_family family
);
179 void ac_get_raster_config(struct radeon_info
*info
,
180 uint32_t *raster_config_p
,
181 uint32_t *raster_config_1_p
,
182 uint32_t *se_tile_repeat_p
);
183 void ac_get_harvested_configs(struct radeon_info
*info
,
184 unsigned raster_config
,
185 unsigned *cik_raster_config_1_p
,
186 unsigned *raster_config_se
);
187 unsigned ac_get_compute_resource_limits(struct radeon_info
*info
,
188 unsigned waves_per_threadgroup
,
189 unsigned max_waves_per_sh
,
190 unsigned threadgroups_per_cu
);
192 static inline unsigned ac_get_max_wave64_per_simd(enum radeon_family family
)
196 /* These always have 8 waves: */
207 static inline unsigned ac_get_num_physical_vgprs(enum chip_class chip_class
,
210 /* The number is per SIMD. */
211 if (chip_class
>= GFX10
)
212 return wave_size
== 32 ? 1024 : 512;
217 static inline uint32_t
218 ac_get_num_physical_sgprs(const struct radeon_info
*info
)
220 /* The number is per SIMD. There is enough SGPRs for the maximum number
221 * of Wave32, which is double the number for Wave64.
223 if (info
->chip_class
>= GFX10
)
224 return 128 * ac_get_max_wave64_per_simd(info
->family
) * 2;
226 return info
->chip_class
>= GFX8
? 800 : 512;
233 #endif /* AC_GPU_INFO_H */