/*
- Copyright (C) Intel Corp. 2006. All Rights Reserved.
- Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
- develop this 3D driver.
-
- Permission is hereby granted, free of charge, to any person obtaining
- a copy of this software and associated documentation files (the
- "Software"), to deal in the Software without restriction, including
- without limitation the rights to use, copy, modify, merge, publish,
- distribute, sublicense, and/or sell copies of the Software, and to
- permit persons to whom the Software is furnished to do so, subject to
- the following conditions:
-
- The above copyright notice and this permission notice (including the
- next paragraph) shall be included in all copies or substantial
- portions of the Software.
-
- THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
- IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
- LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
- OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
- WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
- **********************************************************************/
- /*
- * Authors:
- * Keith Whitwell <keith@tungstengraphics.com>
- */
-
-#include "main/glheader.h"
-#include "main/macros.h"
-#include "main/enums.h"
-#include "main/transformfeedback.h"
-
-#include "intel_batchbuffer.h"
-
-#include "brw_defines.h"
+ * Copyright © 2013 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file brw_vec4_gs.c
+ *
+ * State atom for client-programmable geometry shaders, and support code.
+ */
+
+#include "brw_gs.h"
#include "brw_context.h"
-#include "brw_eu.h"
-#include "brw_util.h"
#include "brw_state.h"
-#include "brw_gs.h"
+#include "brw_ff_gs.h"
+#include "compiler/brw_nir.h"
+#include "brw_program.h"
+#include "compiler/glsl/ir_uniform.h"
+
+static void
+brw_gs_debug_recompile(struct brw_context *brw, struct gl_program *prog,
+ const struct brw_gs_prog_key *key)
+{
+ perf_debug("Recompiling geometry shader for program %d\n", prog->Id);
+
+ bool found = false;
+ const struct brw_gs_prog_key *old_key =
+ brw_find_previous_compile(&brw->cache, BRW_CACHE_GS_PROG,
+ key->program_string_id);
+
+ if (!old_key) {
+ perf_debug(" Didn't find previous compile in the shader cache for "
+ "debug\n");
+ return;
+ }
-#include "glsl/ralloc.h"
+ found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
-static void compile_gs_prog( struct brw_context *brw,
- struct brw_gs_prog_key *key )
+ if (!found) {
+ perf_debug(" Something else\n");
+ }
+}
+
+static void
+assign_gs_binding_table_offsets(const struct gen_device_info *devinfo,
+ const struct gl_program *prog,
+ struct brw_gs_prog_data *prog_data)
{
- struct intel_context *intel = &brw->intel;
- struct brw_gs_compile c;
- const GLuint *program;
- void *mem_ctx;
- GLuint program_size;
-
- memset(&c, 0, sizeof(c));
-
- c.key = *key;
- c.vue_map = brw->vs.prog_data->base.vue_map;
- c.nr_regs = (c.vue_map.num_slots + 1)/2;
-
- mem_ctx = ralloc_context(NULL);
-
- /* Begin the compilation:
+ /* In gen6 we reserve the first BRW_MAX_SOL_BINDINGS entries for transform
+ * feedback surfaces.
*/
- brw_init_compile(brw, &c.func, mem_ctx);
+ uint32_t reserved = devinfo->gen == 6 ? BRW_MAX_SOL_BINDINGS : 0;
- c.func.single_program_flow = 1;
+ brw_assign_common_binding_table_offsets(devinfo, prog,
+ &prog_data->base.base, reserved);
+}
+
+static bool
+brw_codegen_gs_prog(struct brw_context *brw,
+ struct brw_program *gp,
+ struct brw_gs_prog_key *key)
+{
+ struct brw_compiler *compiler = brw->screen->compiler;
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
+ struct brw_stage_state *stage_state = &brw->gs.base;
+ struct brw_gs_prog_data prog_data;
+ bool start_busy = false;
+ double start_time = 0;
+
+ memset(&prog_data, 0, sizeof(prog_data));
- /* For some reason the thread is spawned with only 4 channels
- * unmasked.
+ assign_gs_binding_table_offsets(devinfo, &gp->program, &prog_data);
+
+ /* Allocate the references to the uniforms that will end up in the
+ * prog_data associated with the compiled program, and which will be freed
+ * by the state cache.
+ *
+ * Note: param_count needs to be num_uniform_components * 4, since we add
+ * padding around uniform values below vec4 size, so the worst case is that
+ * every uniform is a float which gets padded to the size of a vec4.
*/
- brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
+ int param_count = gp->program.nir->num_uniforms / 4;
- if (intel->gen >= 6) {
- unsigned num_verts;
- bool check_edge_flag;
- /* On Sandybridge, we use the GS for implementing transform feedback
- * (called "Stream Out" in the PRM).
- */
- switch (key->primitive) {
- case _3DPRIM_POINTLIST:
- num_verts = 1;
- check_edge_flag = false;
- break;
- case _3DPRIM_LINELIST:
- case _3DPRIM_LINESTRIP:
- case _3DPRIM_LINELOOP:
- num_verts = 2;
- check_edge_flag = false;
- break;
- case _3DPRIM_TRILIST:
- case _3DPRIM_TRIFAN:
- case _3DPRIM_TRISTRIP:
- case _3DPRIM_RECTLIST:
- num_verts = 3;
- check_edge_flag = false;
- break;
- case _3DPRIM_QUADLIST:
- case _3DPRIM_QUADSTRIP:
- case _3DPRIM_POLYGON:
- num_verts = 3;
- check_edge_flag = true;
- break;
- default:
- assert(!"Unexpected primitive type in Gen6 SOL program.");
- return;
+ prog_data.base.base.param =
+ rzalloc_array(NULL, const gl_constant_value *, param_count);
+ prog_data.base.base.pull_param =
+ rzalloc_array(NULL, const gl_constant_value *, param_count);
+ prog_data.base.base.image_param =
+ rzalloc_array(NULL, struct brw_image_param,
+ gp->program.info.num_images);
+ prog_data.base.base.nr_params = param_count;
+ prog_data.base.base.nr_image_params = gp->program.info.num_images;
+
+ brw_nir_setup_glsl_uniforms(gp->program.nir, &gp->program,
+ &prog_data.base.base,
+ compiler->scalar_stage[MESA_SHADER_GEOMETRY]);
+
+ uint64_t outputs_written = gp->program.info.outputs_written;
+
+ brw_compute_vue_map(devinfo,
+ &prog_data.base.vue_map, outputs_written,
+ gp->program.info.separate_shader);
+
+ int st_index = -1;
+ if (INTEL_DEBUG & DEBUG_SHADER_TIME)
+ st_index = brw_get_shader_time_index(brw, &gp->program, ST_GS, true);
+
+ if (unlikely(brw->perf_debug)) {
+ start_busy = brw->batch.last_bo && brw_bo_busy(brw->batch.last_bo);
+ start_time = get_time();
+ }
+
+ void *mem_ctx = ralloc_context(NULL);
+ unsigned program_size;
+ char *error_str;
+ const unsigned *program =
+ brw_compile_gs(brw->screen->compiler, brw, mem_ctx, key,
+ &prog_data, gp->program.nir, &gp->program,
+ st_index, &program_size, &error_str);
+ if (program == NULL) {
+ ralloc_strcat(&gp->program.sh.data->InfoLog, error_str);
+ _mesa_problem(NULL, "Failed to compile geometry shader: %s\n", error_str);
+
+ ralloc_free(mem_ctx);
+ return false;
+ }
+
+ if (unlikely(brw->perf_debug)) {
+ if (gp->compiled_once) {
+ brw_gs_debug_recompile(brw, &gp->program, key);
}
- gen6_sol_program(&c, key, num_verts, check_edge_flag);
- } else {
- /* On Gen4-5, we use the GS to decompose certain types of primitives.
- * Note that primitives which don't require a GS program have already
- * been weeded out by now.
- */
- switch (key->primitive) {
- case _3DPRIM_QUADLIST:
- brw_gs_quads( &c, key );
- break;
- case _3DPRIM_QUADSTRIP:
- brw_gs_quad_strip( &c, key );
- break;
- case _3DPRIM_LINELOOP:
- brw_gs_lines( &c );
- break;
- default:
- ralloc_free(mem_ctx);
- return;
+ if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
+ perf_debug("GS compile took %.03f ms and stalled the GPU\n",
+ (get_time() - start_time) * 1000);
}
+ gp->compiled_once = true;
}
- /* get the program
- */
- program = brw_get_program(&c.func, &program_size);
-
- if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
- int i;
-
- printf("gs:\n");
- for (i = 0; i < program_size / sizeof(struct brw_instruction); i++)
- brw_disasm(stdout, &((struct brw_instruction *)program)[i],
- intel->gen);
- printf("\n");
- }
-
- brw_upload_cache(&brw->cache, BRW_GS_PROG,
- &c.key, sizeof(c.key),
- program, program_size,
- &c.prog_data, sizeof(c.prog_data),
- &brw->gs.prog_offset, &brw->gs.prog_data);
+ /* Scratch space is used for register spilling */
+ brw_alloc_stage_scratch(brw, stage_state,
+ prog_data.base.base.total_scratch,
+ devinfo->max_gs_threads);
+
+ brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
+ key, sizeof(*key),
+ program, program_size,
+ &prog_data, sizeof(prog_data),
+ &stage_state->prog_offset, &brw->gs.base.prog_data);
ralloc_free(mem_ctx);
+
+ return true;
}
-static void populate_key( struct brw_context *brw,
- struct brw_gs_prog_key *key )
+static bool
+brw_gs_state_dirty(const struct brw_context *brw)
{
- static const unsigned swizzle_for_offset[4] = {
- BRW_SWIZZLE4(0, 1, 2, 3),
- BRW_SWIZZLE4(1, 2, 3, 3),
- BRW_SWIZZLE4(2, 3, 3, 3),
- BRW_SWIZZLE4(3, 3, 3, 3)
- };
+ return brw_state_dirty(brw,
+ _NEW_TEXTURE,
+ BRW_NEW_GEOMETRY_PROGRAM |
+ BRW_NEW_TRANSFORM_FEEDBACK);
+}
- struct gl_context *ctx = &brw->intel.ctx;
- struct intel_context *intel = &brw->intel;
+void
+brw_gs_populate_key(struct brw_context *brw,
+ struct brw_gs_prog_key *key)
+{
+ struct gl_context *ctx = &brw->ctx;
+ struct brw_program *gp = (struct brw_program *) brw->geometry_program;
memset(key, 0, sizeof(*key));
- /* CACHE_NEW_VS_PROG (part of VUE map) */
- key->attrs = brw->vs.prog_data->base.vue_map.slots_valid;
+ key->program_string_id = gp->id;
+
+ /* _NEW_TEXTURE */
+ brw_populate_sampler_prog_key_data(ctx, &gp->program, &key->tex);
+}
- /* BRW_NEW_PRIMITIVE */
- key->primitive = brw->primitive;
+void
+brw_upload_gs_prog(struct brw_context *brw)
+{
+ struct brw_stage_state *stage_state = &brw->gs.base;
+ struct brw_gs_prog_key key;
+ /* BRW_NEW_GEOMETRY_PROGRAM */
+ struct brw_program *gp = (struct brw_program *) brw->geometry_program;
- /* _NEW_LIGHT */
- key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
- if (key->primitive == _3DPRIM_QUADLIST && ctx->Light.ShadeModel != GL_FLAT) {
- /* Provide consistent primitive order with brw_set_prim's
- * optimization of single quads to trifans.
- */
- key->pv_first = true;
- }
+ if (!brw_gs_state_dirty(brw))
+ return;
- if (intel->gen >= 7) {
- /* On Gen7 and later, we don't use GS (yet). */
- key->need_gs_prog = false;
- } else if (intel->gen == 6) {
- /* On Gen6, GS is used for transform feedback. */
- /* _NEW_TRANSFORM_FEEDBACK */
- if (_mesa_is_xfb_active_and_unpaused(ctx)) {
- const struct gl_shader_program *shaderprog =
- ctx->Shader.CurrentVertexProgram;
- const struct gl_transform_feedback_info *linked_xfb_info =
- &shaderprog->LinkedTransformFeedback;
- int i;
-
- /* Make sure that the VUE slots won't overflow the unsigned chars in
- * key->transform_feedback_bindings[].
- */
- STATIC_ASSERT(BRW_VARYING_SLOT_COUNT <= 256);
-
- /* Make sure that we don't need more binding table entries than we've
- * set aside for use in transform feedback. (We shouldn't, since we
- * set aside enough binding table entries to have one per component).
- */
- assert(linked_xfb_info->NumOutputs <= BRW_MAX_SOL_BINDINGS);
-
- key->need_gs_prog = true;
- key->num_transform_feedback_bindings = linked_xfb_info->NumOutputs;
- for (i = 0; i < key->num_transform_feedback_bindings; ++i) {
- key->transform_feedback_bindings[i] =
- linked_xfb_info->Outputs[i].OutputRegister;
- key->transform_feedback_swizzles[i] =
- swizzle_for_offset[linked_xfb_info->Outputs[i].ComponentOffset];
- }
+ if (gp == NULL) {
+ /* No geometry shader. Vertex data just passes straight through. */
+ if (brw->gen == 6 &&
+ (brw->ctx.NewDriverState & BRW_NEW_TRANSFORM_FEEDBACK)) {
+ gen6_brw_upload_ff_gs_prog(brw);
+ return;
}
- /* On Gen6, GS is also used for rasterizer discard. */
- /* _NEW_RASTERIZER_DISCARD */
- if (ctx->RasterDiscard) {
- key->need_gs_prog = true;
- key->rasterizer_discard = true;
- }
- } else {
- /* Pre-gen6, GS is used to transform QUADLIST, QUADSTRIP, and LINELOOP
- * into simpler primitives.
+
+ /* Other state atoms had better not try to access prog_data, since
+ * there's no GS program.
*/
- key->need_gs_prog = (brw->primitive == _3DPRIM_QUADLIST ||
- brw->primitive == _3DPRIM_QUADSTRIP ||
- brw->primitive == _3DPRIM_LINELOOP);
+ brw->gs.base.prog_data = NULL;
+
+ return;
+ }
+
+ brw_gs_populate_key(brw, &key);
+
+ if (!brw_search_cache(&brw->cache, BRW_CACHE_GS_PROG,
+ &key, sizeof(key),
+ &stage_state->prog_offset,
+ &brw->gs.base.prog_data)) {
+ bool success = brw_codegen_gs_prog(brw, gp, &key);
+ assert(success);
+ (void)success;
}
}
-/* Calculate interpolants for triangle and line rasterization.
- */
-static void
-brw_upload_gs_prog(struct brw_context *brw)
+bool
+brw_gs_precompile(struct gl_context *ctx, struct gl_program *prog)
{
+ struct brw_context *brw = brw_context(ctx);
struct brw_gs_prog_key key;
- /* Populate the key:
- */
- populate_key(brw, &key);
+ uint32_t old_prog_offset = brw->gs.base.prog_offset;
+ struct brw_stage_prog_data *old_prog_data = brw->gs.base.prog_data;
+ bool success;
- if (brw->gs.prog_active != key.need_gs_prog) {
- brw->state.dirty.cache |= CACHE_NEW_GS_PROG;
- brw->gs.prog_active = key.need_gs_prog;
- }
+ struct brw_program *bgp = brw_program(prog);
- if (brw->gs.prog_active) {
- if (!brw_search_cache(&brw->cache, BRW_GS_PROG,
- &key, sizeof(key),
- &brw->gs.prog_offset, &brw->gs.prog_data)) {
- compile_gs_prog( brw, &key );
- }
- }
-}
+ memset(&key, 0, sizeof(key));
+ brw_setup_tex_for_precompile(brw, &key.tex, prog);
+ key.program_string_id = bgp->id;
-const struct brw_tracked_state brw_gs_prog = {
- .dirty = {
- .mesa = (_NEW_LIGHT |
- _NEW_TRANSFORM_FEEDBACK |
- _NEW_RASTERIZER_DISCARD),
- .brw = BRW_NEW_PRIMITIVE,
- .cache = CACHE_NEW_VS_PROG
- },
- .emit = brw_upload_gs_prog
-};
+ success = brw_codegen_gs_prog(brw, bgp, &key);
+
+ brw->gs.base.prog_offset = old_prog_offset;
+ brw->gs.base.prog_data = old_prog_data;
+
+ return success;
+}