4dbba0174891bb99836e6bfdaaaeaf287622b28a
[mesa.git] / src / intel / blorp / blorp.c
1 /*
2 * Copyright © 2012 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * IN THE SOFTWARE.
22 */
23
24 #include <errno.h>
25
26 #include "program/prog_instruction.h"
27
28 #include "blorp_priv.h"
29 #include "brw_compiler.h"
30 #include "brw_nir.h"
31
32 void
33 blorp_init(struct blorp_context *blorp, void *driver_ctx,
34 struct isl_device *isl_dev)
35 {
36 blorp->driver_ctx = driver_ctx;
37 blorp->isl_dev = isl_dev;
38 }
39
40 void
41 blorp_finish(struct blorp_context *blorp)
42 {
43 blorp->driver_ctx = NULL;
44 }
45
46 void
47 blorp_batch_init(struct blorp_context *blorp,
48 struct blorp_batch *batch, void *driver_batch)
49 {
50 batch->blorp = blorp;
51 batch->driver_batch = driver_batch;
52 }
53
54 void
55 blorp_batch_finish(struct blorp_batch *batch)
56 {
57 batch->blorp = NULL;
58 }
59
60 void
61 brw_blorp_surface_info_init(struct blorp_context *blorp,
62 struct brw_blorp_surface_info *info,
63 const struct blorp_surf *surf,
64 unsigned int level, unsigned int layer,
65 enum isl_format format, bool is_render_target)
66 {
67 /* Layer is a physical layer, so if this is a 2D multisample array texture
68 * using INTEL_MSAA_LAYOUT_UMS or INTEL_MSAA_LAYOUT_CMS, then it had better
69 * be a multiple of num_samples.
70 */
71 unsigned layer_multiplier = 1;
72 if (surf->surf->msaa_layout == ISL_MSAA_LAYOUT_ARRAY) {
73 assert(layer % surf->surf->samples == 0);
74 layer_multiplier = surf->surf->samples;
75 }
76
77 if (format == ISL_FORMAT_UNSUPPORTED)
78 format = surf->surf->format;
79
80 if (format == ISL_FORMAT_R24_UNORM_X8_TYPELESS) {
81 /* Unfortunately, ISL_FORMAT_R24_UNORM_X8_TYPELESS it isn't supported as
82 * a render target, which would prevent us from blitting to 24-bit
83 * depth. The miptree consists of 32 bits per pixel, arranged as 24-bit
84 * depth values interleaved with 8 "don't care" bits. Since depth
85 * values don't require any blending, it doesn't matter how we interpret
86 * the bit pattern as long as we copy the right amount of data, so just
87 * map it as 8-bit BGRA.
88 */
89 format = ISL_FORMAT_B8G8R8A8_UNORM;
90 } else if (surf->surf->usage & ISL_SURF_USAGE_STENCIL_BIT) {
91 assert(surf->surf->format == ISL_FORMAT_R8_UINT);
92 /* Prior to Broadwell, we can't render to R8_UINT */
93 if (blorp->isl_dev->info->gen < 8)
94 format = ISL_FORMAT_R8_UNORM;
95 }
96
97 info->surf = *surf->surf;
98 info->addr = surf->addr;
99
100 info->aux_usage = surf->aux_usage;
101 if (info->aux_usage != ISL_AUX_USAGE_NONE) {
102 info->aux_surf = *surf->aux_surf;
103 info->aux_addr = surf->aux_addr;
104 }
105
106 info->clear_color = surf->clear_color;
107
108 info->view = (struct isl_view) {
109 .usage = is_render_target ? ISL_SURF_USAGE_RENDER_TARGET_BIT :
110 ISL_SURF_USAGE_TEXTURE_BIT,
111 .format = format,
112 .base_level = level,
113 .levels = 1,
114 .channel_select = {
115 ISL_CHANNEL_SELECT_RED,
116 ISL_CHANNEL_SELECT_GREEN,
117 ISL_CHANNEL_SELECT_BLUE,
118 ISL_CHANNEL_SELECT_ALPHA,
119 },
120 };
121
122 if (!is_render_target &&
123 (info->surf.dim == ISL_SURF_DIM_3D ||
124 info->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY)) {
125 /* 3-D textures don't support base_array layer and neither do 2-D
126 * multisampled textures on IVB so we need to pass it through the
127 * sampler in those cases. These are also two cases where we are
128 * guaranteed that we won't be doing any funny surface hacks.
129 */
130 info->view.base_array_layer = 0;
131 info->view.array_len = MAX2(info->surf.logical_level0_px.depth,
132 info->surf.logical_level0_px.array_len);
133 info->z_offset = layer / layer_multiplier;
134 } else {
135 info->view.base_array_layer = layer / layer_multiplier;
136 info->view.array_len = 1;
137 info->z_offset = 0;
138 }
139 }
140
141
142 void
143 blorp_params_init(struct blorp_params *params)
144 {
145 memset(params, 0, sizeof(*params));
146 params->num_draw_buffers = 1;
147 params->num_layers = 1;
148 }
149
150 void
151 brw_blorp_init_wm_prog_key(struct brw_wm_prog_key *wm_key)
152 {
153 memset(wm_key, 0, sizeof(*wm_key));
154 wm_key->nr_color_regions = 1;
155 for (int i = 0; i < MAX_SAMPLERS; i++)
156 wm_key->tex.swizzles[i] = SWIZZLE_XYZW;
157 }
158
159 static int
160 nir_uniform_type_size(const struct glsl_type *type)
161 {
162 /* Only very basic types are allowed */
163 assert(glsl_type_is_vector_or_scalar(type));
164 assert(glsl_get_bit_size(type) == 32);
165
166 return glsl_get_vector_elements(type) * 4;
167 }
168
169 const unsigned *
170 brw_blorp_compile_nir_shader(struct blorp_context *blorp, struct nir_shader *nir,
171 const struct brw_wm_prog_key *wm_key,
172 bool use_repclear,
173 struct brw_blorp_prog_data *prog_data,
174 unsigned *program_size)
175 {
176 const struct brw_compiler *compiler = blorp->compiler;
177
178 void *mem_ctx = ralloc_context(NULL);
179
180 /* Calling brw_preprocess_nir and friends is destructive and, if cloning is
181 * enabled, may end up completely replacing the nir_shader. Therefore, we
182 * own it and might as well put it in our context for easy cleanup.
183 */
184 ralloc_steal(mem_ctx, nir);
185 nir->options =
186 compiler->glsl_compiler_options[MESA_SHADER_FRAGMENT].NirOptions;
187
188 struct brw_wm_prog_data wm_prog_data;
189 memset(&wm_prog_data, 0, sizeof(wm_prog_data));
190
191 wm_prog_data.base.nr_params = 0;
192 wm_prog_data.base.param = NULL;
193
194 /* BLORP always just uses the first two binding table entries */
195 wm_prog_data.binding_table.render_target_start = BLORP_RENDERBUFFER_BT_INDEX;
196 wm_prog_data.base.binding_table.texture_start = BLORP_TEXTURE_BT_INDEX;
197
198 nir = brw_preprocess_nir(compiler, nir);
199 nir_remove_dead_variables(nir, nir_var_shader_in);
200 nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
201
202 /* Uniforms are required to be lowered before going into compile_fs. For
203 * BLORP, we'll assume that whoever builds the shader sets the location
204 * they want so we just need to lower them and figure out how many we have
205 * in total.
206 */
207 nir->num_uniforms = 0;
208 nir_foreach_variable(var, &nir->uniforms) {
209 var->data.driver_location = var->data.location;
210 unsigned end = var->data.location + nir_uniform_type_size(var->type);
211 nir->num_uniforms = MAX2(nir->num_uniforms, end);
212 }
213 nir_lower_io(nir, nir_var_uniform, nir_uniform_type_size);
214
215 const unsigned *program =
216 brw_compile_fs(compiler, blorp->driver_ctx, mem_ctx,
217 wm_key, &wm_prog_data, nir,
218 NULL, -1, -1, false, use_repclear, program_size, NULL);
219
220 /* Copy the relavent bits of wm_prog_data over into the blorp prog data */
221 prog_data->dispatch_8 = wm_prog_data.dispatch_8;
222 prog_data->dispatch_16 = wm_prog_data.dispatch_16;
223 prog_data->first_curbe_grf_0 = wm_prog_data.base.dispatch_grf_start_reg;
224 prog_data->first_curbe_grf_2 = wm_prog_data.dispatch_grf_start_reg_2;
225 prog_data->ksp_offset_2 = wm_prog_data.prog_offset_2;
226 prog_data->persample_msaa_dispatch = wm_prog_data.persample_dispatch;
227 prog_data->flat_inputs = wm_prog_data.flat_inputs;
228 prog_data->num_varying_inputs = wm_prog_data.num_varying_inputs;
229 prog_data->inputs_read = nir->info.inputs_read;
230
231 assert(wm_prog_data.base.nr_params == 0);
232
233 return program;
234 }
235
236 void
237 blorp_gen6_hiz_op(struct blorp_batch *batch,
238 struct blorp_surf *surf, unsigned level, unsigned layer,
239 enum blorp_hiz_op op)
240 {
241 struct blorp_params params;
242 blorp_params_init(&params);
243
244 params.hiz_op = op;
245
246 brw_blorp_surface_info_init(batch->blorp, &params.depth, surf, level, layer,
247 surf->surf->format, true);
248
249 /* Align the rectangle primitive to 8x4 pixels.
250 *
251 * During fast depth clears, the emitted rectangle primitive must be
252 * aligned to 8x4 pixels. From the Ivybridge PRM, Vol 2 Part 1 Section
253 * 11.5.3.1 Depth Buffer Clear (and the matching section in the Sandybridge
254 * PRM):
255 * If Number of Multisamples is NUMSAMPLES_1, the rectangle must be
256 * aligned to an 8x4 pixel block relative to the upper left corner
257 * of the depth buffer [...]
258 *
259 * For hiz resolves, the rectangle must also be 8x4 aligned. Item
260 * WaHizAmbiguate8x4Aligned from the Haswell workarounds page and the
261 * Ivybridge simulator require the alignment.
262 *
263 * To be safe, let's just align the rect for all hiz operations and all
264 * hardware generations.
265 *
266 * However, for some miptree slices of a Z24 texture, emitting an 8x4
267 * aligned rectangle that covers the slice may clobber adjacent slices if
268 * we strictly adhered to the texture alignments specified in the PRM. The
269 * Ivybridge PRM, Section "Alignment Unit Size", states that
270 * SURFACE_STATE.Surface_Horizontal_Alignment should be 4 for Z24 surfaces,
271 * not 8. But commit 1f112cc increased the alignment from 4 to 8, which
272 * prevents the clobbering.
273 */
274 params.x1 = minify(params.depth.surf.logical_level0_px.width,
275 params.depth.view.base_level);
276 params.y1 = minify(params.depth.surf.logical_level0_px.height,
277 params.depth.view.base_level);
278 params.x1 = ALIGN(params.x1, 8);
279 params.y1 = ALIGN(params.y1, 4);
280
281 if (params.depth.view.base_level == 0) {
282 /* TODO: What about MSAA? */
283 params.depth.surf.logical_level0_px.width = params.x1;
284 params.depth.surf.logical_level0_px.height = params.y1;
285 }
286
287 params.dst.surf.samples = params.depth.surf.samples;
288 params.dst.surf.logical_level0_px = params.depth.surf.logical_level0_px;
289 params.depth_format = isl_format_get_depth_format(surf->surf->format, false);
290
291 batch->blorp->exec(batch, &params);
292 }