1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
4 * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
26 * Rob Clark <robclark@freedesktop.org>
32 #include "pipe/p_state.h"
33 #include "compiler/shader_enums.h"
34 #include "util/bitscan.h"
41 /* driver param indices: */
42 enum ir3_driver_param
{
43 /* compute shader driver params: */
44 IR3_DP_NUM_WORK_GROUPS_X
= 0,
45 IR3_DP_NUM_WORK_GROUPS_Y
= 1,
46 IR3_DP_NUM_WORK_GROUPS_Z
= 2,
47 IR3_DP_CS_COUNT
= 4, /* must be aligned to vec4 */
49 /* vertex shader driver params: */
50 IR3_DP_VTXID_BASE
= 0,
51 IR3_DP_VTXCNT_MAX
= 1,
52 /* user-clip-plane components, up to 8x vec4's: */
56 IR3_DP_VS_COUNT
= 36 /* must be aligned to vec4 */
60 * For consts needed to pass internal values to shader which may or may not
61 * be required, rather than allocating worst-case const space, we scan the
62 * shader and allocate consts as-needed:
64 * + SSBO sizes: only needed if shader has a get_buffer_size intrinsic
67 struct ir3_driver_const_layout
{
69 uint32_t mask
; /* bitmask of SSBOs that have get_buffer_size */
70 uint32_t count
; /* number of consts allocated */
71 /* one const allocated per SSBO which has get_buffer_size,
72 * ssbo_sizes.off[ssbo_id] is offset from start of ssbo_sizes
75 uint32_t off
[PIPE_MAX_SHADER_BUFFERS
];
79 /* Configuration key used to identify a shader variant.. different
80 * shader variants can be used to implement features not supported
81 * in hw (two sided color), binning-pass vertex shader, etc.
83 struct ir3_shader_key
{
87 * Combined Vertex/Fragment shader parameters:
89 unsigned ucp_enables
: 8;
91 /* do we need to check {v,f}saturate_{s,t,r}? */
92 unsigned has_per_samp
: 1;
95 * Vertex shader variant parameters:
97 unsigned binning_pass
: 1;
98 unsigned vclamp_color
: 1;
101 * Fragment shader variant parameters:
103 unsigned color_two_side
: 1;
104 unsigned half_precision
: 1;
105 /* used when shader needs to handle flat varyings (a4xx)
106 * for front/back color inputs to frag shader:
108 unsigned rasterflat
: 1;
109 unsigned fclamp_color
: 1;
114 /* bitmask of sampler which needs coords clamped for vertex
117 uint16_t vsaturate_s
, vsaturate_t
, vsaturate_r
;
119 /* bitmask of sampler which needs coords clamped for frag
122 uint16_t fsaturate_s
, fsaturate_t
, fsaturate_r
;
124 /* bitmask of samplers which need astc srgb workaround: */
125 uint16_t vastc_srgb
, fastc_srgb
;
129 ir3_shader_key_equal(struct ir3_shader_key
*a
, struct ir3_shader_key
*b
)
131 /* slow-path if we need to check {v,f}saturate_{s,t,r} */
132 if (a
->has_per_samp
|| b
->has_per_samp
)
133 return memcmp(a
, b
, sizeof(struct ir3_shader_key
)) == 0;
134 return a
->global
== b
->global
;
137 /* will the two keys produce different lowering for a fragment shader? */
139 ir3_shader_key_changes_fs(struct ir3_shader_key
*key
, struct ir3_shader_key
*last_key
)
141 if (last_key
->has_per_samp
|| key
->has_per_samp
) {
142 if ((last_key
->fsaturate_s
!= key
->fsaturate_s
) ||
143 (last_key
->fsaturate_t
!= key
->fsaturate_t
) ||
144 (last_key
->fsaturate_r
!= key
->fsaturate_r
) ||
145 (last_key
->fastc_srgb
!= key
->fastc_srgb
))
149 if (last_key
->fclamp_color
!= key
->fclamp_color
)
152 if (last_key
->color_two_side
!= key
->color_two_side
)
155 if (last_key
->half_precision
!= key
->half_precision
)
158 if (last_key
->rasterflat
!= key
->rasterflat
)
161 if (last_key
->ucp_enables
!= key
->ucp_enables
)
167 /* will the two keys produce different lowering for a vertex shader? */
169 ir3_shader_key_changes_vs(struct ir3_shader_key
*key
, struct ir3_shader_key
*last_key
)
171 if (last_key
->has_per_samp
|| key
->has_per_samp
) {
172 if ((last_key
->vsaturate_s
!= key
->vsaturate_s
) ||
173 (last_key
->vsaturate_t
!= key
->vsaturate_t
) ||
174 (last_key
->vsaturate_r
!= key
->vsaturate_r
) ||
175 (last_key
->vastc_srgb
!= key
->vastc_srgb
))
179 if (last_key
->vclamp_color
!= key
->vclamp_color
)
182 if (last_key
->ucp_enables
!= key
->ucp_enables
)
188 struct ir3_shader_variant
{
191 /* variant id (for debug) */
194 struct ir3_shader_key key
;
196 struct ir3_driver_const_layout const_layout
;
197 struct ir3_info info
;
200 /* the instructions length is in units of instruction groups
201 * (4 instructions for a3xx, 16 instructions for a4xx.. each
202 * instruction is 2 dwords):
206 /* the constants length is in units of vec4's, and is the sum of
207 * the uniforms and the built-in compiler constants
211 /* number of uniforms (in vec4), not including built-in compiler
214 unsigned num_uniforms
;
219 * + Let the frag shader determine the position/compmask for the
220 * varyings, since it is the place where we know if the varying
221 * is actually used, and if so, which components are used. So
222 * what the hw calls "outloc" is taken from the "inloc" of the
224 * + From the vert shader, we only need the output regid
227 /* for frag shader, pos_regid holds the frag_pos, ie. what is passed
228 * to bary.f instructions
231 bool frag_coord
, frag_face
, color0_mrt
;
233 /* NOTE: for input/outputs, slot is:
234 * gl_vert_attrib - for VS inputs
235 * gl_varying_slot - for VS output / FS input
236 * gl_frag_result - for FS output
239 /* varyings/outputs: */
240 unsigned outputs_count
;
244 } outputs
[16 + 2]; /* +POSITION +PSIZE */
245 bool writes_pos
, writes_psize
;
247 /* attributes (VS) / varyings (FS):
248 * Note that sysval's should come *after* normal inputs.
250 unsigned inputs_count
;
256 /* location of input (ie. offset passed to bary.f, etc). This
257 * matches the SP_VS_VPC_DST_REG.OUTLOCn value (a3xx and a4xx
258 * have the OUTLOCn value offset by 8, presumably to account
259 * for gl_Position/gl_PointSize)
262 /* vertex shader specific: */
263 bool sysval
: 1; /* slot is a gl_system_value */
264 /* fragment shader specific: */
265 bool bary
: 1; /* fetched varying (vs one loaded into reg) */
266 bool rasterflat
: 1; /* special handling for emit->rasterflat */
267 enum glsl_interp_mode interpolate
;
268 } inputs
[16 + 2]; /* +POSITION +FACE */
270 /* sum of input components (scalar). For frag shaders, it only counts
271 * the varying inputs:
275 /* For frag shaders, the total number of inputs (not scalar,
276 * ie. SP_VS_PARAM_REG.TOTALVSOUTVAR)
280 /* do we have one or more texture sample instructions: */
283 /* do we have one or more SSBO instructions: */
286 /* do we have kill instructions: */
289 /* Layout of constant registers, each section (in vec4). Pointer size
290 * is 32b (a3xx, a4xx), or 64b (a5xx+), which effects the size of the
291 * UBO and stream-out consts.
294 /* user const start at zero */
296 /* NOTE that a3xx might need a section for SSBO addresses too */
298 unsigned driver_param
;
303 unsigned immediates_count
;
308 /* for astc srgb workaround, the number/base of additional
309 * alpha tex states we need, and index of original tex states
312 unsigned base
, count
;
313 unsigned orig_idx
[16];
316 /* shader variants form a linked list: */
317 struct ir3_shader_variant
*next
;
319 /* replicated here to avoid passing extra ptrs everywhere: */
321 struct ir3_shader
*shader
;
324 typedef struct nir_shader nir_shader
;
329 /* shader id (for debug): */
331 uint32_t variant_count
;
333 /* so we know when we can disable TGSI related hacks: */
336 struct ir3_compiler
*compiler
;
339 struct pipe_stream_output_info stream_output
;
341 struct ir3_shader_variant
*variants
;
344 void * ir3_shader_assemble(struct ir3_shader_variant
*v
, uint32_t gpu_id
);
346 struct ir3_shader
* ir3_shader_create(struct ir3_compiler
*compiler
,
347 const struct pipe_shader_state
*cso
, enum shader_t type
,
348 struct pipe_debug_callback
*debug
);
350 ir3_shader_create_compute(struct ir3_compiler
*compiler
,
351 const struct pipe_compute_state
*cso
,
352 struct pipe_debug_callback
*debug
);
353 void ir3_shader_destroy(struct ir3_shader
*shader
);
354 struct ir3_shader_variant
* ir3_shader_variant(struct ir3_shader
*shader
,
355 struct ir3_shader_key key
, struct pipe_debug_callback
*debug
);
356 void ir3_shader_disasm(struct ir3_shader_variant
*so
, uint32_t *bin
);
357 uint64_t ir3_shader_outputs(const struct ir3_shader
*so
);
359 struct fd_ringbuffer
;
361 void ir3_emit_vs_consts(const struct ir3_shader_variant
*v
, struct fd_ringbuffer
*ring
,
362 struct fd_context
*ctx
, const struct pipe_draw_info
*info
);
363 void ir3_emit_fs_consts(const struct ir3_shader_variant
*v
, struct fd_ringbuffer
*ring
,
364 struct fd_context
*ctx
);
365 void ir3_emit_cs_consts(const struct ir3_shader_variant
*v
, struct fd_ringbuffer
*ring
,
366 struct fd_context
*ctx
, const struct pipe_grid_info
*info
);
369 ir3_glsl_type_size(const struct glsl_type
*type
);
371 static inline const char *
372 ir3_shader_stage(struct ir3_shader
*shader
)
374 switch (shader
->type
) {
375 case SHADER_VERTEX
: return "VERT";
376 case SHADER_FRAGMENT
: return "FRAG";
377 case SHADER_COMPUTE
: return "CL";
379 unreachable("invalid type");
388 #include "pipe/p_shader_tokens.h"
391 ir3_find_output(const struct ir3_shader_variant
*so
, gl_varying_slot slot
)
395 for (j
= 0; j
< so
->outputs_count
; j
++)
396 if (so
->outputs
[j
].slot
== slot
)
399 /* it seems optional to have a OUT.BCOLOR[n] for each OUT.COLOR[n]
400 * in the vertex shader.. but the fragment shader doesn't know this
401 * so it will always have both IN.COLOR[n] and IN.BCOLOR[n]. So
402 * at link time if there is no matching OUT.BCOLOR[n], we must map
403 * OUT.COLOR[n] to IN.BCOLOR[n]. And visa versa if there is only
404 * a OUT.BCOLOR[n] but no matching OUT.COLOR[n]
406 if (slot
== VARYING_SLOT_BFC0
) {
407 slot
= VARYING_SLOT_COL0
;
408 } else if (slot
== VARYING_SLOT_BFC1
) {
409 slot
= VARYING_SLOT_COL1
;
410 } else if (slot
== VARYING_SLOT_COL0
) {
411 slot
= VARYING_SLOT_BFC0
;
412 } else if (slot
== VARYING_SLOT_COL1
) {
413 slot
= VARYING_SLOT_BFC1
;
418 for (j
= 0; j
< so
->outputs_count
; j
++)
419 if (so
->outputs
[j
].slot
== slot
)
428 ir3_next_varying(const struct ir3_shader_variant
*so
, int i
)
430 while (++i
< so
->inputs_count
)
431 if (so
->inputs
[i
].compmask
&& so
->inputs
[i
].bary
)
436 struct ir3_shader_linkage
{
447 ir3_link_add(struct ir3_shader_linkage
*l
, uint8_t regid
, uint8_t compmask
, uint8_t loc
)
451 debug_assert(i
< ARRAY_SIZE(l
->var
));
453 l
->var
[i
].regid
= regid
;
454 l
->var
[i
].compmask
= compmask
;
456 l
->max_loc
= MAX2(l
->max_loc
, loc
+ util_last_bit(compmask
));
460 ir3_link_shaders(struct ir3_shader_linkage
*l
,
461 const struct ir3_shader_variant
*vs
,
462 const struct ir3_shader_variant
*fs
)
466 while (l
->cnt
< ARRAY_SIZE(l
->var
)) {
467 j
= ir3_next_varying(fs
, j
);
469 if (j
>= fs
->inputs_count
)
472 if (fs
->inputs
[j
].inloc
>= fs
->total_in
)
475 k
= ir3_find_output(vs
, fs
->inputs
[j
].slot
);
477 ir3_link_add(l
, vs
->outputs
[k
].regid
,
478 fs
->inputs
[j
].compmask
, fs
->inputs
[j
].inloc
);
482 static inline uint32_t
483 ir3_find_output_regid(const struct ir3_shader_variant
*so
, unsigned slot
)
486 for (j
= 0; j
< so
->outputs_count
; j
++)
487 if (so
->outputs
[j
].slot
== slot
)
488 return so
->outputs
[j
].regid
;
492 static inline uint32_t
493 ir3_find_sysval_regid(const struct ir3_shader_variant
*so
, unsigned slot
)
496 for (j
= 0; j
< so
->inputs_count
; j
++)
497 if (so
->inputs
[j
].sysval
&& (so
->inputs
[j
].slot
== slot
))
498 return so
->inputs
[j
].regid
;
502 #endif /* IR3_SHADER_H_ */