2 * Copyright © 2014 Broadcom
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
34 #include "util/macros.h"
35 #include "glsl/nir/nir.h"
36 #include "util/list.h"
37 #include "util/u_math.h"
39 #include "vc4_screen.h"
40 #include "pipe/p_state.h"
52 * Stores an immediate value in the index field that can be turned
53 * into a small immediate field by qpu_encode_small_immediate().
87 /* Note: Orderings of these compares must be the same as in
88 * qpu_defines.h. Selects the src[0] if the ns flag bit is set,
94 /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
113 QOP_TLB_DISCARD_SETUP,
114 QOP_TLB_STENCIL_SETUP,
140 /** Texture x coordinate parameter write */
142 /** Texture y coordinate parameter write */
144 /** Texture border color parameter or cube map z coordinate write */
146 /** Texture LOD bias parameter write */
150 * Texture-unit 4-byte read with address provided direct in S
153 * The first operand is the offset from the start of the UBO, and the
154 * second is the uniform that has the UBO's base pointer.
159 * Signal of texture read being necessary and then reading r4 into
165 struct queued_qpu_inst {
166 struct list_head link;
171 struct list_head link;
181 * Coordinate shader, runs during binning, before the VS, and just
189 enum quniform_contents {
191 * Indicates that a constant 32-bit value is copied from the program's
196 * Indicates that the program's uniform contents are used as an index
197 * into the GL uniform storage.
202 * Scaling factors from clip coordinates to relative to the viewport
205 * This is used by the coordinate and vertex shaders to produce the
206 * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
207 * point offsets from the viewport ccenter.
209 QUNIFORM_VIEWPORT_X_SCALE,
210 QUNIFORM_VIEWPORT_Y_SCALE,
213 QUNIFORM_VIEWPORT_Z_OFFSET,
214 QUNIFORM_VIEWPORT_Z_SCALE,
216 QUNIFORM_USER_CLIP_PLANE,
219 * A reference to a texture config parameter 0 uniform.
221 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
222 * defines texture type, miplevels, and such. It will be found as a
223 * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
225 QUNIFORM_TEXTURE_CONFIG_P0,
228 * A reference to a texture config parameter 1 uniform.
230 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
231 * defines texture width, height, filters, and wrap modes. It will be
232 * found as a parameter to the second QOP_TEX_[STRB] instruction in a
235 QUNIFORM_TEXTURE_CONFIG_P1,
237 /** A reference to a texture config parameter 2 cubemap stride uniform */
238 QUNIFORM_TEXTURE_CONFIG_P2,
242 QUNIFORM_TEXRECT_SCALE_X,
243 QUNIFORM_TEXRECT_SCALE_Y,
245 QUNIFORM_TEXTURE_BORDER_COLOR,
247 QUNIFORM_BLEND_CONST_COLOR_X,
248 QUNIFORM_BLEND_CONST_COLOR_Y,
249 QUNIFORM_BLEND_CONST_COLOR_Z,
250 QUNIFORM_BLEND_CONST_COLOR_W,
257 struct vc4_varying_semantic {
263 struct vc4_compiler_ubo_range {
265 * offset in bytes from the start of the ubo where this range is
268 * Only set once used is set.
273 * offset in bytes from the start of the gallium uniforms where the
278 /** size in bytes of this ubo range */
282 * Set if this range is used by the shader for indirect uniforms
289 struct vc4_uncompiled_shader *shader_state;
291 enum pipe_format format;
292 unsigned compare_mode:1;
293 unsigned compare_func:3;
297 } tex[VC4_MAX_TEXTURE_SAMPLERS];
303 enum pipe_format color_format;
305 bool stencil_enabled;
306 bool stencil_twoside;
307 bool stencil_full_writemasks;
311 bool point_coord_upper_left;
313 uint8_t alpha_test_func;
314 uint8_t logicop_func;
315 uint32_t point_sprite_mask;
317 struct pipe_rt_blend_state blend;
324 * This is a proxy for the array of FS input semantics, which is
325 * larger than we would want to put in the key.
327 uint64_t compiled_fs_id;
329 enum pipe_format attr_formats[8];
331 bool per_vertex_point_size;
335 struct vc4_context *vc4;
337 nir_function_impl *impl;
338 struct exec_list *cf_node_list;
341 * Mapping from nir_register * or nir_ssa_def * to array of struct
342 * qreg for the values.
344 struct hash_table *def_ht;
346 /* For each temp, the instruction generating its value. */
348 uint32_t defs_array_size;
351 * Inputs to the shader, arranged by TGSI declaration order.
353 * Not all fragment shader QFILE_VARY reads are present in this array.
356 struct qreg *outputs;
357 uint32_t inputs_array_size;
358 uint32_t outputs_array_size;
359 uint32_t uniforms_array_size;
361 struct vc4_compiler_ubo_range *ubo_ranges;
362 uint32_t ubo_ranges_array_size;
363 /** Number of uniform areas declared in ubo_ranges. */
364 uint32_t num_uniform_ranges;
365 /** Number of uniform areas used for indirect addressed loads. */
366 uint32_t num_ubo_ranges;
367 uint32_t next_ubo_dst_offset;
369 struct qreg line_x, point_x, point_y;
372 uint8_t vattr_sizes[8];
375 * Array of the TGSI semantics of all FS QFILE_VARY reads.
377 * This includes those that aren't part of the VPM varyings, like
378 * point/line coordinates.
380 struct vc4_varying_semantic *input_semantics;
381 uint32_t num_input_semantics;
382 uint32_t input_semantics_array_size;
385 * An entry per outputs[] in the VS indicating what the semantic of
386 * the output is. Used to emit from the VS in the order that the FS
389 struct vc4_varying_semantic *output_semantics;
391 struct pipe_shader_state *shader_state;
393 struct vc4_fs_key *fs_key;
394 struct vc4_vs_key *vs_key;
396 uint32_t *uniform_data;
397 enum quniform_contents *uniform_contents;
398 uint32_t uniform_array_size;
399 uint32_t num_uniforms;
400 uint32_t num_outputs;
401 uint32_t num_texture_samples;
402 uint32_t output_position_index;
403 uint32_t output_clipvertex_index;
404 uint32_t output_color_index;
405 uint32_t output_point_size_index;
410 struct list_head instructions;
411 uint32_t immediates[1024];
413 struct list_head qpu_inst_list;
415 uint32_t qpu_inst_count;
416 uint32_t qpu_inst_size;
423 /* Special nir_load_input intrinsic index for loading the current TLB
426 #define VC4_NIR_TLB_COLOR_READ_INPUT 2000000000
428 /* Special offset for nir_load_uniform values to get a QUNIFORM_*
429 * state-dependent value.
431 #define VC4_NIR_STATE_UNIFORM_OFFSET 2000000000
433 struct vc4_compile *qir_compile_init(void);
434 void qir_compile_destroy(struct vc4_compile *c);
435 struct qinst *qir_inst(enum qop op, struct qreg dst,
436 struct qreg src0, struct qreg src1);
437 struct qinst *qir_inst4(enum qop op, struct qreg dst,
442 void qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst);
443 struct qreg qir_uniform(struct vc4_compile *c,
444 enum quniform_contents contents,
446 void qir_reorder_uniforms(struct vc4_compile *c);
448 void qir_emit(struct vc4_compile *c, struct qinst *inst);
449 static inline void qir_emit_nodef(struct vc4_compile *c, struct qinst *inst)
451 list_addtail(&inst->link, &c->instructions);
454 struct qreg qir_get_temp(struct vc4_compile *c);
455 int qir_get_op_nsrc(enum qop qop);
456 bool qir_reg_equals(struct qreg a, struct qreg b);
457 bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
458 bool qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst);
459 bool qir_is_multi_instruction(struct qinst *inst);
460 bool qir_is_mul(struct qinst *inst);
461 bool qir_is_tex(struct qinst *inst);
462 bool qir_depends_on_flags(struct qinst *inst);
463 bool qir_writes_r4(struct qinst *inst);
464 bool qir_src_needs_a_file(struct qinst *inst);
465 struct qreg qir_follow_movs(struct vc4_compile *c, struct qreg reg);
467 void qir_dump(struct vc4_compile *c);
468 void qir_dump_inst(struct vc4_compile *c, struct qinst *inst);
469 const char *qir_get_stage_name(enum qstage stage);
471 void qir_optimize(struct vc4_compile *c);
472 bool qir_opt_algebraic(struct vc4_compile *c);
473 bool qir_opt_constant_folding(struct vc4_compile *c);
474 bool qir_opt_copy_propagation(struct vc4_compile *c);
475 bool qir_opt_cse(struct vc4_compile *c);
476 bool qir_opt_dead_code(struct vc4_compile *c);
477 bool qir_opt_small_immediates(struct vc4_compile *c);
478 bool qir_opt_vpm_writes(struct vc4_compile *c);
479 void vc4_nir_lower_blend(struct vc4_compile *c);
480 void vc4_nir_lower_io(struct vc4_compile *c);
481 nir_ssa_def *vc4_nir_get_state_uniform(struct nir_builder *b,
482 enum quniform_contents contents);
483 nir_ssa_def *vc4_nir_get_swizzled_channel(struct nir_builder *b,
484 nir_ssa_def **srcs, int swiz);
485 void qir_lower_uniforms(struct vc4_compile *c);
487 void qpu_schedule_instructions(struct vc4_compile *c);
489 void qir_SF(struct vc4_compile *c, struct qreg src);
491 static inline struct qreg
492 qir_uniform_ui(struct vc4_compile *c, uint32_t ui)
494 return qir_uniform(c, QUNIFORM_CONSTANT, ui);
497 static inline struct qreg
498 qir_uniform_f(struct vc4_compile *c, float f)
500 return qir_uniform(c, QUNIFORM_CONSTANT, fui(f));
503 #define QIR_ALU0(name) \
504 static inline struct qreg \
505 qir_##name(struct vc4_compile *c) \
507 struct qreg t = qir_get_temp(c); \
508 qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef)); \
512 #define QIR_ALU1(name) \
513 static inline struct qreg \
514 qir_##name(struct vc4_compile *c, struct qreg a) \
516 struct qreg t = qir_get_temp(c); \
517 qir_emit(c, qir_inst(QOP_##name, t, a, c->undef)); \
521 qir_##name##_dest(struct vc4_compile *c, struct qreg dest, \
524 qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, c->undef)); \
527 #define QIR_ALU2(name) \
528 static inline struct qreg \
529 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b) \
531 struct qreg t = qir_get_temp(c); \
532 qir_emit(c, qir_inst(QOP_##name, t, a, b)); \
536 qir_##name##_dest(struct vc4_compile *c, struct qreg dest, \
537 struct qreg a, struct qreg b) \
539 qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, b)); \
542 #define QIR_NODST_1(name) \
544 qir_##name(struct vc4_compile *c, struct qreg a) \
546 qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef)); \
549 #define QIR_NODST_2(name) \
551 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b) \
553 qir_emit(c, qir_inst(QOP_##name, c->undef, a, b)); \
556 #define QIR_PACK(name) \
557 static inline struct qreg \
558 qir_##name(struct vc4_compile *c, struct qreg dest, struct qreg a) \
560 qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, c->undef)); \
600 QIR_ALU1(PACK_8888_F)
610 QIR_NODST_2(TEX_DIRECT)
615 QIR_ALU0(FRAG_REV_FLAG)
617 QIR_ALU0(TLB_COLOR_READ)
618 QIR_NODST_1(TLB_COLOR_WRITE)
619 QIR_NODST_1(TLB_Z_WRITE)
620 QIR_NODST_1(TLB_DISCARD_SETUP)
621 QIR_NODST_1(TLB_STENCIL_SETUP)
623 static inline struct qreg
624 qir_UNPACK_8_F(struct vc4_compile *c, struct qreg src, int i)
626 struct qreg t = qir_get_temp(c);
627 qir_emit(c, qir_inst(QOP_UNPACK_8A_F + i, t, src, c->undef));
631 static inline struct qreg
632 qir_UNPACK_8_I(struct vc4_compile *c, struct qreg src, int i)
634 struct qreg t = qir_get_temp(c);
635 qir_emit(c, qir_inst(QOP_UNPACK_8A_I + i, t, src, c->undef));
639 static inline struct qreg
640 qir_UNPACK_16_F(struct vc4_compile *c, struct qreg src, int i)
642 struct qreg t = qir_get_temp(c);
643 qir_emit(c, qir_inst(QOP_UNPACK_16A_F + i, t, src, c->undef));
647 static inline struct qreg
648 qir_UNPACK_16_I(struct vc4_compile *c, struct qreg src, int i)
650 struct qreg t = qir_get_temp(c);
651 qir_emit(c, qir_inst(QOP_UNPACK_16A_I + i, t, src, c->undef));
655 static inline struct qreg
656 qir_PACK_8_F(struct vc4_compile *c, struct qreg dest, struct qreg val, int chan)
658 qir_emit(c, qir_inst(QOP_PACK_8A_F + chan, dest, val, c->undef));
659 if (dest.file == QFILE_TEMP)
660 c->defs[dest.index] = NULL;
664 static inline struct qreg
665 qir_POW(struct vc4_compile *c, struct qreg x, struct qreg y)
667 return qir_EXP2(c, qir_FMUL(c,
673 qir_VPM_WRITE(struct vc4_compile *c, struct qreg val)
675 static const struct qreg vpm = { QFILE_VPM, 0 };
676 qir_emit(c, qir_inst(QOP_MOV, vpm, val, c->undef));
679 #endif /* VC4_QIR_H */