#include "main/compiler.h"
+#include "main/context.h"
#include "brw_context.h"
#include "brw_vs.h"
#include "brw_util.h"
#include "brw_state.h"
#include "program/prog_print.h"
#include "program/prog_parameter.h"
-#include "brw_nir.h"
+#include "compiler/brw_nir.h"
+#include "brw_program.h"
#include "util/ralloc.h"
* Decide which set of clip planes should be used when clipping via
* gl_Position or gl_ClipVertex.
*/
-gl_clip_plane *brw_select_clip_planes(struct gl_context *ctx)
+gl_clip_plane *
+brw_select_clip_planes(struct gl_context *ctx)
{
if (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]) {
/* There is currently a GLSL vertex shader, so clip according to GLSL
}
}
-bool
-brw_codegen_vs_prog(struct brw_context *brw,
- struct gl_shader_program *prog,
- struct brw_vertex_program *vp,
- struct brw_vs_prog_key *key)
+GLbitfield64
+brw_vs_outputs_written(struct brw_context *brw, struct brw_vs_prog_key *key,
+ GLbitfield64 user_varyings)
{
- GLuint program_size;
- const GLuint *program;
- struct brw_vs_prog_data prog_data;
- struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
- void *mem_ctx;
- int i;
- struct brw_shader *vs = NULL;
- bool start_busy = false;
- double start_time = 0;
-
- if (!vp->program.Base.nir) {
- /* Normally we generate NIR in LinkShader() or
- * ProgramStringNotify(), but Mesa's fixed-function vertex program
- * handling doesn't notify the driver at all. Just do it here, at
- * the last minute, even though it's lame.
- */
- assert(vp->program.Base.Id == 0 && prog == NULL);
- vp->program.Base.nir =
- brw_create_nir(brw, NULL, &vp->program.Base, MESA_SHADER_VERTEX,
- brw->intelScreen->compiler->scalar_vs);
- }
-
- if (prog)
- vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
-
- memset(&prog_data, 0, sizeof(prog_data));
-
- /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
- if (!prog)
- stage_prog_data->use_alt_mode = true;
-
- mem_ctx = ralloc_context(NULL);
-
- /* 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.
- */
- int param_count = vp->program.Base.nir->num_uniforms;
- if (!brw->intelScreen->compiler->scalar_vs)
- param_count *= 4;
-
- if (vs)
- prog_data.base.base.nr_image_params = vs->base.NumImages;
-
- /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
- * planes as uniforms.
- */
- param_count += key->nr_userclip_plane_consts * 4;
-
- stage_prog_data->param =
- rzalloc_array(NULL, const gl_constant_value *, param_count);
- stage_prog_data->pull_param =
- rzalloc_array(NULL, const gl_constant_value *, param_count);
- stage_prog_data->image_param =
- rzalloc_array(NULL, struct brw_image_param,
- stage_prog_data->nr_image_params);
- stage_prog_data->nr_params = param_count;
-
- GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
- prog_data.inputs_read = vp->program.Base.InputsRead;
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
+ GLbitfield64 outputs_written = user_varyings;
if (key->copy_edgeflag) {
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
- prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
}
- if (brw->gen < 6) {
+ if (devinfo->gen < 6) {
/* Put dummy slots into the VUE for the SF to put the replaced
* point sprite coords in. We shouldn't need these dummy slots,
* which take up precious URB space, but it would mean that the SF
* doesn't get nice aligned pairs of input coords into output
* coords, which would be a pain to handle.
*/
- for (i = 0; i < 8; i++) {
+ for (unsigned i = 0; i < 8; i++) {
if (key->point_coord_replace & (1 << i))
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
}
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
}
- brw_compute_vue_map(brw->intelScreen->devinfo,
- &prog_data.base.vue_map, outputs_written,
- prog ? prog->SeparateShader : false);
-
- if (0) {
- _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
- true);
- }
-
- if (unlikely(brw->perf_debug)) {
- start_busy = (brw->batch.last_bo &&
- drm_intel_bo_busy(brw->batch.last_bo));
- start_time = get_time();
- }
-
- /* Emit GEN4 code.
- */
- program = brw_vs_emit(brw, mem_ctx, key, &prog_data,
- &vp->program, prog, &program_size);
- if (program == NULL) {
- ralloc_free(mem_ctx);
- return false;
- }
-
- if (unlikely(brw->perf_debug) && vs) {
- if (vs->compiled_once) {
- brw_vs_debug_recompile(brw, prog, key);
- }
- if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
- perf_debug("VS compile took %.03f ms and stalled the GPU\n",
- (get_time() - start_time) * 1000);
- }
- vs->compiled_once = true;
- }
-
- /* Scratch space is used for register spilling */
- if (prog_data.base.base.total_scratch) {
- brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
- prog_data.base.base.total_scratch *
- brw->max_vs_threads);
- }
-
- brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
- key, sizeof(struct brw_vs_prog_key),
- program, program_size,
- &prog_data, sizeof(prog_data),
- &brw->vs.base.prog_offset, &brw->vs.prog_data);
- ralloc_free(mem_ctx);
-
- return true;
+ return outputs_written;
}
-static bool
-key_debug(struct brw_context *brw, const char *name, int a, int b)
-{
- if (a != b) {
- perf_debug(" %s %d->%d\n", name, a, b);
- return true;
- }
- return false;
-}
-
-void
-brw_vs_debug_recompile(struct brw_context *brw,
- struct gl_shader_program *prog,
+static void
+brw_vs_debug_recompile(struct brw_context *brw, struct gl_program *prog,
const struct brw_vs_prog_key *key)
{
- struct brw_cache_item *c = NULL;
- const struct brw_vs_prog_key *old_key = NULL;
- bool found = false;
+ perf_debug("Recompiling vertex shader for program %d\n", prog->Id);
- perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
-
- for (unsigned int i = 0; i < brw->cache.size; i++) {
- for (c = brw->cache.items[i]; c; c = c->next) {
- if (c->cache_id == BRW_CACHE_VS_PROG) {
- old_key = c->key;
-
- if (old_key->program_string_id == key->program_string_id)
- break;
- }
- }
- if (c)
- break;
- }
+ bool found = false;
+ const struct brw_vs_prog_key *old_key =
+ brw_find_previous_compile(&brw->cache, BRW_CACHE_VS_PROG,
+ key->program_string_id);
- if (!c) {
+ if (!old_key) {
perf_debug(" Didn't find previous compile in the shader cache for "
"debug\n");
return;
}
static bool
-brw_vs_state_dirty(struct brw_context *brw)
+brw_codegen_vs_prog(struct brw_context *brw,
+ struct brw_program *vp,
+ struct brw_vs_prog_key *key)
+{
+ const struct brw_compiler *compiler = brw->screen->compiler;
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
+ const GLuint *program;
+ struct brw_vs_prog_data prog_data;
+ struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
+ void *mem_ctx;
+ bool start_busy = false;
+ double start_time = 0;
+
+ memset(&prog_data, 0, sizeof(prog_data));
+
+ /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
+ if (vp->program.is_arb_asm)
+ stage_prog_data->use_alt_mode = true;
+
+ mem_ctx = ralloc_context(NULL);
+
+ brw_assign_common_binding_table_offsets(devinfo, &vp->program,
+ &prog_data.base.base, 0);
+
+ if (!vp->program.is_arb_asm) {
+ brw_nir_setup_glsl_uniforms(mem_ctx, vp->program.nir, &vp->program,
+ &prog_data.base.base,
+ compiler->scalar_stage[MESA_SHADER_VERTEX]);
+ brw_nir_analyze_ubo_ranges(compiler, vp->program.nir,
+ prog_data.base.base.ubo_ranges);
+ } else {
+ brw_nir_setup_arb_uniforms(mem_ctx, vp->program.nir, &vp->program,
+ &prog_data.base.base);
+ }
+
+ uint64_t outputs_written =
+ brw_vs_outputs_written(brw, key, vp->program.nir->info.outputs_written);
+
+ brw_compute_vue_map(devinfo,
+ &prog_data.base.vue_map, outputs_written,
+ vp->program.nir->info.separate_shader);
+
+ if (0) {
+ _mesa_fprint_program_opt(stderr, &vp->program, PROG_PRINT_DEBUG, true);
+ }
+
+ if (unlikely(brw->perf_debug)) {
+ start_busy = (brw->batch.last_bo &&
+ brw_bo_busy(brw->batch.last_bo));
+ start_time = get_time();
+ }
+
+ if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
+ if (vp->program.is_arb_asm)
+ brw_dump_arb_asm("vertex", &vp->program);
+ }
+
+ int st_index = -1;
+ if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
+ st_index = brw_get_shader_time_index(brw, &vp->program, ST_VS,
+ !vp->program.is_arb_asm);
+ }
+
+ /* Emit GEN4 code.
+ */
+ char *error_str;
+ program = brw_compile_vs(compiler, brw, mem_ctx, key, &prog_data,
+ vp->program.nir,
+ st_index, &error_str);
+ if (program == NULL) {
+ if (!vp->program.is_arb_asm) {
+ vp->program.sh.data->LinkStatus = LINKING_FAILURE;
+ ralloc_strcat(&vp->program.sh.data->InfoLog, error_str);
+ }
+
+ _mesa_problem(NULL, "Failed to compile vertex shader: %s\n", error_str);
+
+ ralloc_free(mem_ctx);
+ return false;
+ }
+
+ if (unlikely(brw->perf_debug)) {
+ if (vp->compiled_once) {
+ brw_vs_debug_recompile(brw, &vp->program, key);
+ }
+ if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
+ perf_debug("VS compile took %.03f ms and stalled the GPU\n",
+ (get_time() - start_time) * 1000);
+ }
+ vp->compiled_once = true;
+ }
+
+ /* Scratch space is used for register spilling */
+ brw_alloc_stage_scratch(brw, &brw->vs.base,
+ prog_data.base.base.total_scratch);
+
+ /* The param and pull_param arrays will be freed by the shader cache. */
+ ralloc_steal(NULL, prog_data.base.base.param);
+ ralloc_steal(NULL, prog_data.base.base.pull_param);
+ brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
+ key, sizeof(struct brw_vs_prog_key),
+ program, prog_data.base.base.program_size,
+ &prog_data, sizeof(prog_data),
+ &brw->vs.base.prog_offset, &brw->vs.base.prog_data);
+ ralloc_free(mem_ctx);
+
+ return true;
+}
+
+static bool
+brw_vs_state_dirty(const struct brw_context *brw)
{
return brw_state_dirty(brw,
_NEW_BUFFERS |
BRW_NEW_VS_ATTRIB_WORKAROUNDS);
}
-static void
+void
brw_vs_populate_key(struct brw_context *brw,
struct brw_vs_prog_key *key)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_VERTEX_PROGRAM */
- struct brw_vertex_program *vp =
- (struct brw_vertex_program *)brw->vertex_program;
- struct gl_program *prog = (struct gl_program *) brw->vertex_program;
- int i;
+ struct gl_program *prog = brw->programs[MESA_SHADER_VERTEX];
+ struct brw_program *vp = (struct brw_program *) prog;
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
memset(key, 0, sizeof(*key));
key->program_string_id = vp->id;
if (ctx->Transform.ClipPlanesEnabled != 0 &&
- ctx->API == API_OPENGL_COMPAT &&
- !vp->program.Base.UsesClipDistanceOut) {
+ (ctx->API == API_OPENGL_COMPAT || ctx->API == API_OPENGLES) &&
+ vp->program.info.clip_distance_array_size == 0) {
key->nr_userclip_plane_consts =
_mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
}
- /* _NEW_POLYGON */
- if (brw->gen < 6) {
+ if (devinfo->gen < 6) {
+ /* _NEW_POLYGON */
key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
ctx->Polygon.BackMode != GL_FILL);
+
+ /* _NEW_POINT */
+ if (ctx->Point.PointSprite) {
+ key->point_coord_replace = ctx->Point.CoordReplace & 0xff;
+ }
}
- if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
- VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
+ if (prog->info.outputs_written &
+ (VARYING_BIT_COL0 | VARYING_BIT_COL1 | VARYING_BIT_BFC0 |
+ VARYING_BIT_BFC1)) {
/* _NEW_LIGHT | _NEW_BUFFERS */
key->clamp_vertex_color = ctx->Light._ClampVertexColor;
}
- /* _NEW_POINT */
- if (brw->gen < 6 && ctx->Point.PointSprite) {
- for (i = 0; i < 8; i++) {
- if (ctx->Point.CoordReplace[i])
- key->point_coord_replace |= (1 << i);
- }
- }
-
/* _NEW_TEXTURE */
- brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
- &key->tex);
+ brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
/* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
- memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
- sizeof(brw->vb.attrib_wa_flags));
+ if (devinfo->gen < 8 && !devinfo->is_haswell) {
+ memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
+ sizeof(brw->vb.attrib_wa_flags));
+ }
}
void
brw_upload_vs_prog(struct brw_context *brw)
{
- struct gl_context *ctx = &brw->ctx;
- struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
struct brw_vs_prog_key key;
/* BRW_NEW_VERTEX_PROGRAM */
- struct brw_vertex_program *vp =
- (struct brw_vertex_program *)brw->vertex_program;
+ struct brw_program *vp =
+ (struct brw_program *) brw->programs[MESA_SHADER_VERTEX];
if (!brw_vs_state_dirty(brw))
return;
brw_vs_populate_key(brw, &key);
- if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
- &key, sizeof(key),
- &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
- bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
- vp, &key);
- (void) success;
- assert(success);
- }
- brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
+ if (brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
+ &key, sizeof(key),
+ &brw->vs.base.prog_offset, &brw->vs.base.prog_data))
+ return;
+
+ if (brw_disk_cache_upload_program(brw, MESA_SHADER_VERTEX))
+ return;
+
+ vp = (struct brw_program *) brw->programs[MESA_SHADER_VERTEX];
+ vp->id = key.program_string_id;
+
+ MAYBE_UNUSED bool success = brw_codegen_vs_prog(brw, vp, &key);
+ assert(success);
}
bool
-brw_vs_precompile(struct gl_context *ctx,
- struct gl_shader_program *shader_prog,
- struct gl_program *prog)
+brw_vs_precompile(struct gl_context *ctx, struct gl_program *prog)
{
struct brw_context *brw = brw_context(ctx);
struct brw_vs_prog_key key;
uint32_t old_prog_offset = brw->vs.base.prog_offset;
- struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
+ struct brw_stage_prog_data *old_prog_data = brw->vs.base.prog_data;
bool success;
- struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
- struct brw_vertex_program *bvp = brw_vertex_program(vp);
+ struct brw_program *bvp = brw_program(prog);
memset(&key, 0, sizeof(key));
brw_setup_tex_for_precompile(brw, &key.tex, prog);
key.program_string_id = bvp->id;
key.clamp_vertex_color =
- (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
- VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
+ (prog->info.outputs_written &
+ (VARYING_BIT_COL0 | VARYING_BIT_COL1 | VARYING_BIT_BFC0 |
+ VARYING_BIT_BFC1));
- success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
+ success = brw_codegen_vs_prog(brw, bvp, &key);
brw->vs.base.prog_offset = old_prog_offset;
- brw->vs.prog_data = old_prog_data;
+ brw->vs.base.prog_data = old_prog_data;
return success;
}