2 * Copyright © 2015 Red Hat
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:
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
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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Rob Clark <robclark@freedesktop.org>
28 #include "nir_builder.h"
30 #define MAX_CLIP_PLANES 8
32 /* Generates the lowering code for user-clip-planes, generating CLIPDIST
33 * from UCP[n] + CLIPVERTEX or POSITION. Additionally, an optional pass
34 * for fragment shaders to insert conditional kills based on the inter-
37 * NOTE: should be run after nir_lower_outputs_to_temporaries() (or at
38 * least in scenarios where you can count on each output written once
44 create_clipdist_var(nir_shader
*shader
, unsigned drvloc
,
45 bool output
, gl_varying_slot slot
, unsigned array_size
)
47 nir_variable
*var
= rzalloc(shader
, nir_variable
);
49 var
->data
.driver_location
= drvloc
;
50 var
->data
.mode
= output
? nir_var_shader_out
: nir_var_shader_in
;
51 var
->name
= ralloc_asprintf(var
, "clipdist_%d", drvloc
);
53 var
->data
.location
= slot
;
56 var
->type
= glsl_array_type(glsl_float_type(), array_size
,
58 var
->data
.compact
= 1;
60 var
->type
= glsl_vec4_type();
63 exec_list_push_tail(&shader
->outputs
, &var
->node
);
64 shader
->num_outputs
++; /* TODO use type_size() */
67 exec_list_push_tail(&shader
->inputs
, &var
->node
);
68 shader
->num_inputs
++; /* TODO use type_size() */
74 create_clipdist_vars(nir_shader
*shader
, nir_variable
**io_vars
,
75 unsigned ucp_enables
, int *drvloc
, bool output
,
76 bool use_clipdist_array
)
78 if (use_clipdist_array
) {
80 create_clipdist_var(shader
, ++(*drvloc
), true,
81 VARYING_SLOT_CLIP_DIST0
,
82 util_last_bit(ucp_enables
));
84 if (ucp_enables
& 0x0f)
86 create_clipdist_var(shader
, ++(*drvloc
), output
,
87 VARYING_SLOT_CLIP_DIST0
, 0);
88 if (ucp_enables
& 0xf0)
90 create_clipdist_var(shader
, ++(*drvloc
), output
,
91 VARYING_SLOT_CLIP_DIST1
, 0);
96 store_clipdist_output(nir_builder
*b
, nir_variable
*out
, nir_ssa_def
**val
)
98 nir_intrinsic_instr
*store
;
100 store
= nir_intrinsic_instr_create(b
->shader
, nir_intrinsic_store_output
);
101 store
->num_components
= 4;
102 nir_intrinsic_set_base(store
, out
->data
.driver_location
);
103 nir_intrinsic_set_write_mask(store
, 0xf);
104 store
->src
[0].ssa
= nir_vec4(b
, val
[0], val
[1], val
[2], val
[3]);
105 store
->src
[0].is_ssa
= true;
106 store
->src
[1] = nir_src_for_ssa(nir_imm_int(b
, 0));
107 nir_builder_instr_insert(b
, &store
->instr
);
111 load_clipdist_input(nir_builder
*b
, nir_variable
*in
, nir_ssa_def
**val
)
113 nir_intrinsic_instr
*load
;
115 load
= nir_intrinsic_instr_create(b
->shader
, nir_intrinsic_load_input
);
116 load
->num_components
= 4;
117 nir_intrinsic_set_base(load
, in
->data
.driver_location
);
118 load
->src
[0] = nir_src_for_ssa(nir_imm_int(b
, 0));
119 nir_ssa_dest_init(&load
->instr
, &load
->dest
, 4, 32, NULL
);
120 nir_builder_instr_insert(b
, &load
->instr
);
122 val
[0] = nir_channel(b
, &load
->dest
.ssa
, 0);
123 val
[1] = nir_channel(b
, &load
->dest
.ssa
, 1);
124 val
[2] = nir_channel(b
, &load
->dest
.ssa
, 2);
125 val
[3] = nir_channel(b
, &load
->dest
.ssa
, 3);
129 find_output_in_block(nir_block
*block
, unsigned drvloc
)
131 nir_foreach_instr(instr
, block
) {
133 if (instr
->type
== nir_instr_type_intrinsic
) {
134 nir_intrinsic_instr
*intr
= nir_instr_as_intrinsic(instr
);
135 if ((intr
->intrinsic
== nir_intrinsic_store_output
) &&
136 nir_intrinsic_base(intr
) == drvloc
) {
137 assert(intr
->src
[0].is_ssa
);
138 assert(nir_src_is_const(intr
->src
[1]));
139 return intr
->src
[0].ssa
;
147 /* TODO: maybe this would be a useful helper?
148 * NOTE: assumes each output is written exactly once (and unconditionally)
149 * so if needed nir_lower_outputs_to_temporaries()
152 find_output(nir_shader
*shader
, unsigned drvloc
)
154 nir_ssa_def
*def
= NULL
;
155 nir_foreach_function(function
, shader
) {
156 if (function
->impl
) {
157 nir_foreach_block_reverse(block
, function
->impl
) {
158 nir_ssa_def
*new_def
= find_output_in_block(block
, drvloc
);
159 assert(!(new_def
&& def
));
162 /* for debug builds, scan entire shader to assert
163 * if output is written multiple times. For release
164 * builds just assume all is well and bail when we
178 find_clipvertex_and_position_outputs(nir_shader
*shader
,
179 nir_variable
**clipvertex
,
180 nir_variable
**position
)
182 nir_foreach_variable(var
, &shader
->outputs
) {
183 switch (var
->data
.location
) {
184 case VARYING_SLOT_POS
:
187 case VARYING_SLOT_CLIP_VERTEX
:
190 case VARYING_SLOT_CLIP_DIST0
:
191 case VARYING_SLOT_CLIP_DIST1
:
192 /* if shader is already writing CLIPDIST, then
193 * there should be no user-clip-planes to deal
196 * We assume nir_remove_dead_variables has removed the clipdist
197 * variables if they're not written.
203 return *clipvertex
|| *position
;
207 get_ucp(nir_builder
*b
, int plane
,
208 const gl_state_index16 clipplane_state_tokens
[][STATE_LENGTH
])
210 if (clipplane_state_tokens
) {
212 snprintf(tmp
, ARRAY_SIZE(tmp
), "gl_ClipPlane%dMESA", plane
);
213 nir_variable
*var
= nir_variable_create(b
->shader
,
218 var
->num_state_slots
= 1;
219 var
->state_slots
= ralloc_array(var
, nir_state_slot
, 1);
220 memcpy(var
->state_slots
[0].tokens
,
221 clipplane_state_tokens
[plane
],
222 sizeof(var
->state_slots
[0].tokens
));
223 return nir_load_var(b
, var
);
225 return nir_load_user_clip_plane(b
, plane
);
230 lower_clip_outputs(nir_builder
*b
, nir_variable
*position
,
231 nir_variable
*clipvertex
, nir_variable
**out
,
232 unsigned ucp_enables
, bool use_vars
,
233 bool use_clipdist_array
,
234 const gl_state_index16 clipplane_state_tokens
[][STATE_LENGTH
])
236 nir_ssa_def
*clipdist
[MAX_CLIP_PLANES
];
240 cv
= nir_load_var(b
, clipvertex
? clipvertex
: position
);
243 exec_node_remove(&clipvertex
->node
);
244 clipvertex
->data
.mode
= nir_var_shader_temp
;
245 exec_list_push_tail(&b
->shader
->globals
, &clipvertex
->node
);
246 nir_fixup_deref_modes(b
->shader
);
250 cv
= find_output(b
->shader
, clipvertex
->data
.driver_location
);
253 cv
= find_output(b
->shader
, position
->data
.driver_location
);
257 for (int plane
= 0; plane
< MAX_CLIP_PLANES
; plane
++) {
258 if (ucp_enables
& (1 << plane
)) {
259 nir_ssa_def
*ucp
= get_ucp(b
, plane
, clipplane_state_tokens
);
261 /* calculate clipdist[plane] - dot(ucp, cv): */
262 clipdist
[plane
] = nir_fdot4(b
, ucp
, cv
);
264 /* 0.0 == don't-clip == disabled: */
265 clipdist
[plane
] = nir_imm_float(b
, 0.0);
267 if (use_clipdist_array
&& plane
< util_last_bit(ucp_enables
)) {
269 nir_deref_instr
*deref
;
270 deref
= nir_build_deref_array_imm(b
,
271 nir_build_deref_var(b
, out
[0]),
273 nir_store_deref(b
, deref
, clipdist
[plane
], 1);
277 if (!use_clipdist_array
) {
279 if (ucp_enables
& 0x0f)
280 nir_store_var(b
, out
[0], nir_vec(b
, clipdist
, 4), 0xf);
281 if (ucp_enables
& 0xf0)
282 nir_store_var(b
, out
[1], nir_vec(b
, &clipdist
[4], 4), 0xf);
284 if (ucp_enables
& 0x0f)
285 store_clipdist_output(b
, out
[0], &clipdist
[0]);
286 if (ucp_enables
& 0xf0)
287 store_clipdist_output(b
, out
[1], &clipdist
[4]);
296 /* ucp_enables is bitmask of enabled ucps. Actual ucp values are
297 * passed in to shader via user_clip_plane system-values
299 * If use_vars is true, the pass will use variable loads and stores instead
300 * of working with store_output intrinsics.
302 * If use_clipdist_array is true, the pass will use compact arrays for the
303 * clipdist output instead of two vec4s.
306 nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
, bool use_vars
,
307 bool use_clipdist_array
,
308 const gl_state_index16 clipplane_state_tokens
[][STATE_LENGTH
])
310 nir_function_impl
*impl
= nir_shader_get_entrypoint(shader
);
313 nir_variable
*position
= NULL
;
314 nir_variable
*clipvertex
= NULL
;
315 nir_variable
*out
[2] = { NULL
};
320 /* find clipvertex/position outputs: */
321 nir_foreach_variable(var
, &shader
->outputs
) {
322 int loc
= var
->data
.driver_location
;
324 /* keep track of last used driver-location.. we'll be
325 * appending CLIP_DIST0/CLIP_DIST1 after last existing
328 maxloc
= MAX2(maxloc
, loc
);
331 nir_builder_init(&b
, impl
);
333 /* NIR should ensure that, even in case of loops/if-else, there
334 * should be only a single predecessor block to end_block, which
335 * makes the perfect place to insert the clipdist calculations.
337 * NOTE: in case of early returns, these would have to be lowered
338 * to jumps to end_block predecessor in a previous pass. Not sure
339 * if there is a good way to sanity check this, but for now the
340 * users of this pass don't support sub-routines.
342 assert(impl
->end_block
->predecessors
->entries
== 1);
343 b
.cursor
= nir_after_cf_list(&impl
->body
);
345 /* find clipvertex/position outputs */
346 if (!find_clipvertex_and_position_outputs(shader
, &clipvertex
, &position
))
349 /* insert CLIPDIST outputs */
350 create_clipdist_vars(shader
, out
, ucp_enables
, &maxloc
, true,
353 lower_clip_outputs(&b
, position
, clipvertex
, out
, ucp_enables
, use_vars
,
354 use_clipdist_array
, clipplane_state_tokens
);
356 nir_metadata_preserve(impl
, nir_metadata_dominance
);
362 lower_clip_in_gs_block(nir_builder
*b
, nir_block
*block
, nir_variable
*position
,
363 nir_variable
*clipvertex
, nir_variable
**out
,
364 unsigned ucp_enables
, bool use_clipdist_array
,
365 const gl_state_index16 clipplane_state_tokens
[][STATE_LENGTH
])
367 nir_foreach_instr_safe(instr
, block
) {
368 if (instr
->type
!= nir_instr_type_intrinsic
)
371 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
372 switch (intrin
->intrinsic
) {
373 case nir_intrinsic_emit_vertex_with_counter
:
374 case nir_intrinsic_emit_vertex
:
375 b
->cursor
= nir_before_instr(instr
);
376 lower_clip_outputs(b
, position
, clipvertex
, out
, ucp_enables
, true,
377 use_clipdist_array
, clipplane_state_tokens
);
380 /* not interesting; skip this */
391 nir_lower_clip_gs(nir_shader
*shader
, unsigned ucp_enables
,
392 bool use_clipdist_array
,
393 const gl_state_index16 clipplane_state_tokens
[][STATE_LENGTH
])
395 nir_function_impl
*impl
= nir_shader_get_entrypoint(shader
);
398 nir_variable
*position
= NULL
;
399 nir_variable
*clipvertex
= NULL
;
400 nir_variable
*out
[2] = { NULL
};
405 /* find clipvertex/position outputs */
406 if (!find_clipvertex_and_position_outputs(shader
, &clipvertex
, &position
))
409 /* insert CLIPDIST outputs */
410 create_clipdist_vars(shader
, out
, ucp_enables
, &maxloc
, true,
413 nir_builder_init(&b
, impl
);
415 nir_foreach_block(block
, impl
)
416 lower_clip_in_gs_block(&b
, block
, position
, clipvertex
, out
,
417 ucp_enables
, use_clipdist_array
,
418 clipplane_state_tokens
);
420 nir_metadata_preserve(impl
, nir_metadata_dominance
);
430 lower_clip_fs(nir_function_impl
*impl
, unsigned ucp_enables
,
433 nir_ssa_def
*clipdist
[MAX_CLIP_PLANES
];
436 nir_builder_init(&b
, impl
);
437 b
.cursor
= nir_before_cf_list(&impl
->body
);
439 if (ucp_enables
& 0x0f)
440 load_clipdist_input(&b
, in
[0], &clipdist
[0]);
441 if (ucp_enables
& 0xf0)
442 load_clipdist_input(&b
, in
[1], &clipdist
[4]);
444 for (int plane
= 0; plane
< MAX_CLIP_PLANES
; plane
++) {
445 if (ucp_enables
& (1 << plane
)) {
446 nir_intrinsic_instr
*discard
;
449 cond
= nir_flt(&b
, clipdist
[plane
], nir_imm_float(&b
, 0.0));
451 discard
= nir_intrinsic_instr_create(b
.shader
,
452 nir_intrinsic_discard_if
);
453 discard
->src
[0] = nir_src_for_ssa(cond
);
454 nir_builder_instr_insert(&b
, &discard
->instr
);
456 b
.shader
->info
.fs
.uses_discard
= true;
460 nir_metadata_preserve(impl
, nir_metadata_dominance
);
463 /* insert conditional kill based on interpolated CLIPDIST
466 nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
,
467 bool use_clipdist_array
)
475 nir_foreach_variable(var
, &shader
->inputs
) {
476 int loc
= var
->data
.driver_location
;
478 /* keep track of last used driver-location.. we'll be
479 * appending CLIP_DIST0/CLIP_DIST1 after last existing
482 maxloc
= MAX2(maxloc
, loc
);
485 /* The shader won't normally have CLIPDIST inputs, so we
488 /* insert CLIPDIST inputs */
489 create_clipdist_vars(shader
, in
, ucp_enables
, &maxloc
, false,
492 nir_foreach_function(function
, shader
) {
493 if (!strcmp(function
->name
, "main"))
494 lower_clip_fs(function
->impl
, ucp_enables
, in
);