1/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
2
3/*
4 * Copyright (C) 2015 Rob Clark <robclark@freedesktop.org>
5 *
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:
12 *
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
15 * Software.
16 *
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
23 * SOFTWARE.
24 *
25 * Authors:
26 *    Rob Clark <robclark@freedesktop.org>
27 */
28
29
30#include "freedreno_util.h"
31
32#include "ir3_nir.h"
33#include "ir3_compiler.h"
34#include "ir3_shader.h"
35
36#include "nir/tgsi_to_nir.h"
37
38static const nir_shader_compiler_options options = {
39		.lower_fpow = true,
40		.lower_fsat = true,
41		.lower_scmp = true,
42		.lower_flrp32 = true,
43		.lower_flrp64 = true,
44		.lower_ffract = true,
45		.lower_fmod32 = true,
46		.lower_fmod64 = true,
47		.lower_fdiv = true,
48		.fuse_ffma = true,
49		.native_integers = true,
50		.vertex_id_zero_based = true,
51		.lower_extract_byte = true,
52		.lower_extract_word = true,
53};
54
55struct nir_shader *
56ir3_tgsi_to_nir(const struct tgsi_token *tokens)
57{
58	return tgsi_to_nir(tokens, &options);
59}
60
61const nir_shader_compiler_options *
62ir3_get_compiler_options(void)
63{
64	return &options;
65}
66
67/* for given shader key, are any steps handled in nir? */
68bool
69ir3_key_lowers_nir(const struct ir3_shader_key *key)
70{
71	return key->fsaturate_s | key->fsaturate_t | key->fsaturate_r |
72			key->vsaturate_s | key->vsaturate_t | key->vsaturate_r |
73			key->ucp_enables | key->color_two_side |
74			key->fclamp_color | key->vclamp_color;
75}
76
77#define OPT(nir, pass, ...) ({                             \
78   bool this_progress = false;                             \
79   NIR_PASS(this_progress, nir, pass, ##__VA_ARGS__);      \
80   this_progress;                                          \
81})
82
83#define OPT_V(nir, pass, ...) NIR_PASS_V(nir, pass, ##__VA_ARGS__)
84
85static void
86ir3_optimize_loop(nir_shader *s)
87{
88	bool progress;
89	do {
90		progress = false;
91
92		OPT_V(s, nir_lower_vars_to_ssa);
93		progress |= OPT(s, nir_lower_alu_to_scalar);
94		progress |= OPT(s, nir_lower_phis_to_scalar);
95
96		progress |= OPT(s, nir_copy_prop);
97		progress |= OPT(s, nir_opt_dce);
98		progress |= OPT(s, nir_opt_cse);
99		progress |= OPT(s, ir3_nir_lower_if_else);
100		progress |= OPT(s, nir_opt_algebraic);
101		progress |= OPT(s, nir_opt_constant_folding);
102
103	} while (progress);
104}
105
106struct nir_shader *
107ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s,
108		const struct ir3_shader_key *key)
109{
110	struct nir_lower_tex_options tex_options = {
111			.lower_rect = 0,
112	};
113
114	if (key) {
115		switch (shader->type) {
116		case SHADER_FRAGMENT:
117		case SHADER_COMPUTE:
118			tex_options.saturate_s = key->fsaturate_s;
119			tex_options.saturate_t = key->fsaturate_t;
120			tex_options.saturate_r = key->fsaturate_r;
121			break;
122		case SHADER_VERTEX:
123			tex_options.saturate_s = key->vsaturate_s;
124			tex_options.saturate_t = key->vsaturate_t;
125			tex_options.saturate_r = key->vsaturate_r;
126			break;
127		}
128	}
129
130	if (shader->compiler->gpu_id >= 400) {
131		/* a4xx seems to have *no* sam.p */
132		tex_options.lower_txp = ~0;  /* lower all txp */
133	} else {
134		/* a3xx just needs to avoid sam.p for 3d tex */
135		tex_options.lower_txp = (1 << GLSL_SAMPLER_DIM_3D);
136	}
137
138	if (fd_mesa_debug & FD_DBG_DISASM) {
139		debug_printf("----------------------\n");
140		nir_print_shader(s, stdout);
141		debug_printf("----------------------\n");
142	}
143
144	OPT_V(s, nir_opt_global_to_local);
145	OPT_V(s, nir_lower_regs_to_ssa);
146
147	if (key) {
148		if (s->stage == MESA_SHADER_VERTEX) {
149			OPT_V(s, nir_lower_clip_vs, key->ucp_enables);
150			if (key->vclamp_color)
151				OPT_V(s, nir_lower_clamp_color_outputs);
152		} else if (s->stage == MESA_SHADER_FRAGMENT) {
153			OPT_V(s, nir_lower_clip_fs, key->ucp_enables);
154			if (key->fclamp_color)
155				OPT_V(s, nir_lower_clamp_color_outputs);
156		}
157		if (key->color_two_side) {
158			OPT_V(s, nir_lower_two_sided_color);
159		}
160	} else {
161		/* only want to do this the first time (when key is null)
162		 * and not again on any potential 2nd variant lowering pass:
163		 */
164		OPT_V(s, ir3_nir_apply_trig_workarounds);
165	}
166
167	OPT_V(s, nir_lower_tex, &tex_options);
168	OPT_V(s, nir_lower_load_const_to_scalar);
169
170	ir3_optimize_loop(s);
171
172	/* do idiv lowering after first opt loop to give a chance for
173	 * divide by immed power-of-two to be caught first:
174	 */
175	if (OPT(s, nir_lower_idiv))
176		ir3_optimize_loop(s);
177
178	OPT_V(s, nir_remove_dead_variables, nir_var_local);
179
180	if (fd_mesa_debug & FD_DBG_DISASM) {
181		debug_printf("----------------------\n");
182		nir_print_shader(s, stdout);
183		debug_printf("----------------------\n");
184	}
185
186	nir_sweep(s);
187
188	return s;
189}
190