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_slot {
262 struct vc4_compiler_ubo_range {
264 * offset in bytes from the start of the ubo where this range is
267 * Only set once used is set.
272 * offset in bytes from the start of the gallium uniforms where the
277 /** size in bytes of this ubo range */
281 * Set if this range is used by the shader for indirect uniforms
288 struct vc4_uncompiled_shader *shader_state;
290 enum pipe_format format;
291 unsigned compare_mode:1;
292 unsigned compare_func:3;
296 } tex[VC4_MAX_TEXTURE_SAMPLERS];
302 enum pipe_format color_format;
304 bool stencil_enabled;
305 bool stencil_twoside;
306 bool stencil_full_writemasks;
310 bool point_coord_upper_left;
312 uint8_t alpha_test_func;
313 uint8_t logicop_func;
314 uint32_t point_sprite_mask;
316 struct pipe_rt_blend_state blend;
323 * This is a proxy for the array of FS input semantics, which is
324 * larger than we would want to put in the key.
326 uint64_t compiled_fs_id;
328 enum pipe_format attr_formats[8];
330 bool per_vertex_point_size;
334 struct vc4_context *vc4;
336 nir_function_impl *impl;
337 struct exec_list *cf_node_list;
340 * Mapping from nir_register * or nir_ssa_def * to array of struct
341 * qreg for the values.
343 struct hash_table *def_ht;
345 /* For each temp, the instruction generating its value. */
347 uint32_t defs_array_size;
350 * Inputs to the shader, arranged by TGSI declaration order.
352 * Not all fragment shader QFILE_VARY reads are present in this array.
355 struct qreg *outputs;
356 uint32_t inputs_array_size;
357 uint32_t outputs_array_size;
358 uint32_t uniforms_array_size;
360 struct vc4_compiler_ubo_range *ubo_ranges;
361 uint32_t ubo_ranges_array_size;
362 /** Number of uniform areas declared in ubo_ranges. */
363 uint32_t num_uniform_ranges;
364 /** Number of uniform areas used for indirect addressed loads. */
365 uint32_t num_ubo_ranges;
366 uint32_t next_ubo_dst_offset;
368 struct qreg line_x, point_x, point_y;
371 uint8_t vattr_sizes[8];
374 * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
376 * This includes those that aren't part of the VPM varyings, like
377 * point/line coordinates.
379 struct vc4_varying_slot *input_slots;
380 uint32_t num_input_slots;
381 uint32_t input_slots_array_size;
384 * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
385 * of the output is. Used to emit from the VS in the order that the
388 struct vc4_varying_slot *output_slots;
390 struct pipe_shader_state *shader_state;
392 struct vc4_fs_key *fs_key;
393 struct vc4_vs_key *vs_key;
395 uint32_t *uniform_data;
396 enum quniform_contents *uniform_contents;
397 uint32_t uniform_array_size;
398 uint32_t num_uniforms;
399 uint32_t num_outputs;
400 uint32_t num_texture_samples;
401 uint32_t output_position_index;
402 uint32_t output_clipvertex_index;
403 uint32_t output_color_index;
404 uint32_t output_point_size_index;
409 struct list_head instructions;
410 uint32_t immediates[1024];
412 struct list_head qpu_inst_list;
414 uint32_t qpu_inst_count;
415 uint32_t qpu_inst_size;
422 /* Special nir_load_input intrinsic index for loading the current TLB
425 #define VC4_NIR_TLB_COLOR_READ_INPUT 2000000000
427 /* Special offset for nir_load_uniform values to get a QUNIFORM_*
428 * state-dependent value.
430 #define VC4_NIR_STATE_UNIFORM_OFFSET 2000000000
432 struct vc4_compile *qir_compile_init(void);
433 void qir_compile_destroy(struct vc4_compile *c);
434 struct qinst *qir_inst(enum qop op, struct qreg dst,
435 struct qreg src0, struct qreg src1);
436 struct qinst *qir_inst4(enum qop op, struct qreg dst,
441 void qir_remove_instruction(struct vc4_compile *c, struct qinst *qinst);
442 struct qreg qir_uniform(struct vc4_compile *c,
443 enum quniform_contents contents,
445 void qir_reorder_uniforms(struct vc4_compile *c);
447 void qir_emit(struct vc4_compile *c, struct qinst *inst);
448 static inline void qir_emit_nodef(struct vc4_compile *c, struct qinst *inst)
450 list_addtail(&inst->link, &c->instructions);
453 struct qreg qir_get_temp(struct vc4_compile *c);
454 int qir_get_op_nsrc(enum qop qop);
455 bool qir_reg_equals(struct qreg a, struct qreg b);
456 bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
457 bool qir_has_side_effect_reads(struct vc4_compile *c, struct qinst *inst);
458 bool qir_is_multi_instruction(struct qinst *inst);
459 bool qir_is_mul(struct qinst *inst);
460 bool qir_is_tex(struct qinst *inst);
461 bool qir_depends_on_flags(struct qinst *inst);
462 bool qir_writes_r4(struct qinst *inst);
463 bool qir_src_needs_a_file(struct qinst *inst);
464 struct qreg qir_follow_movs(struct vc4_compile *c, struct qreg reg);
466 void qir_dump(struct vc4_compile *c);
467 void qir_dump_inst(struct vc4_compile *c, struct qinst *inst);
468 const char *qir_get_stage_name(enum qstage stage);
470 void qir_optimize(struct vc4_compile *c);
471 bool qir_opt_algebraic(struct vc4_compile *c);
472 bool qir_opt_constant_folding(struct vc4_compile *c);
473 bool qir_opt_copy_propagation(struct vc4_compile *c);
474 bool qir_opt_cse(struct vc4_compile *c);
475 bool qir_opt_dead_code(struct vc4_compile *c);
476 bool qir_opt_small_immediates(struct vc4_compile *c);
477 bool qir_opt_vpm_writes(struct vc4_compile *c);
478 void vc4_nir_lower_blend(struct vc4_compile *c);
479 void vc4_nir_lower_io(struct vc4_compile *c);
480 nir_ssa_def *vc4_nir_get_state_uniform(struct nir_builder *b,
481 enum quniform_contents contents);
482 nir_ssa_def *vc4_nir_get_swizzled_channel(struct nir_builder *b,
483 nir_ssa_def **srcs, int swiz);
484 void qir_lower_uniforms(struct vc4_compile *c);
486 void qpu_schedule_instructions(struct vc4_compile *c);
488 void qir_SF(struct vc4_compile *c, struct qreg src);
490 static inline struct qreg
491 qir_uniform_ui(struct vc4_compile *c, uint32_t ui)
493 return qir_uniform(c, QUNIFORM_CONSTANT, ui);
496 static inline struct qreg
497 qir_uniform_f(struct vc4_compile *c, float f)
499 return qir_uniform(c, QUNIFORM_CONSTANT, fui(f));
502 #define QIR_ALU0(name) \
503 static inline struct qreg \
504 qir_##name(struct vc4_compile *c) \
506 struct qreg t = qir_get_temp(c); \
507 qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef)); \
511 #define QIR_ALU1(name) \
512 static inline struct qreg \
513 qir_##name(struct vc4_compile *c, struct qreg a) \
515 struct qreg t = qir_get_temp(c); \
516 qir_emit(c, qir_inst(QOP_##name, t, a, c->undef)); \
520 qir_##name##_dest(struct vc4_compile *c, struct qreg dest, \
523 qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, c->undef)); \
526 #define QIR_ALU2(name) \
527 static inline struct qreg \
528 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b) \
530 struct qreg t = qir_get_temp(c); \
531 qir_emit(c, qir_inst(QOP_##name, t, a, b)); \
535 qir_##name##_dest(struct vc4_compile *c, struct qreg dest, \
536 struct qreg a, struct qreg b) \
538 qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, b)); \
541 #define QIR_NODST_1(name) \
543 qir_##name(struct vc4_compile *c, struct qreg a) \
545 qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef)); \
548 #define QIR_NODST_2(name) \
550 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b) \
552 qir_emit(c, qir_inst(QOP_##name, c->undef, a, b)); \
555 #define QIR_PACK(name) \
556 static inline struct qreg \
557 qir_##name(struct vc4_compile *c, struct qreg dest, struct qreg a) \
559 qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, c->undef)); \
599 QIR_ALU1(PACK_8888_F)
609 QIR_NODST_2(TEX_DIRECT)
614 QIR_ALU0(FRAG_REV_FLAG)
616 QIR_ALU0(TLB_COLOR_READ)
617 QIR_NODST_1(TLB_COLOR_WRITE)
618 QIR_NODST_1(TLB_Z_WRITE)
619 QIR_NODST_1(TLB_DISCARD_SETUP)
620 QIR_NODST_1(TLB_STENCIL_SETUP)
622 static inline struct qreg
623 qir_UNPACK_8_F(struct vc4_compile *c, struct qreg src, int i)
625 struct qreg t = qir_get_temp(c);
626 qir_emit(c, qir_inst(QOP_UNPACK_8A_F + i, t, src, c->undef));
630 static inline struct qreg
631 qir_UNPACK_8_I(struct vc4_compile *c, struct qreg src, int i)
633 struct qreg t = qir_get_temp(c);
634 qir_emit(c, qir_inst(QOP_UNPACK_8A_I + i, t, src, c->undef));
638 static inline struct qreg
639 qir_UNPACK_16_F(struct vc4_compile *c, struct qreg src, int i)
641 struct qreg t = qir_get_temp(c);
642 qir_emit(c, qir_inst(QOP_UNPACK_16A_F + i, t, src, c->undef));
646 static inline struct qreg
647 qir_UNPACK_16_I(struct vc4_compile *c, struct qreg src, int i)
649 struct qreg t = qir_get_temp(c);
650 qir_emit(c, qir_inst(QOP_UNPACK_16A_I + i, t, src, c->undef));
654 static inline struct qreg
655 qir_PACK_8_F(struct vc4_compile *c, struct qreg dest, struct qreg val, int chan)
657 qir_emit(c, qir_inst(QOP_PACK_8A_F + chan, dest, val, c->undef));
658 if (dest.file == QFILE_TEMP)
659 c->defs[dest.index] = NULL;
663 static inline struct qreg
664 qir_POW(struct vc4_compile *c, struct qreg x, struct qreg y)
666 return qir_EXP2(c, qir_FMUL(c,
672 qir_VPM_WRITE(struct vc4_compile *c, struct qreg val)
674 static const struct qreg vpm = { QFILE_VPM, 0 };
675 qir_emit(c, qir_inst(QOP_MOV, vpm, val, c->undef));
678 #endif /* VC4_QIR_H */