/**************************************************************************
- *
+ *
* Copyright 2007 VMware, Inc.
* All Rights Reserved.
- *
+ *
* 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
* distribute, sub license, 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 NON-INFRINGEMENT.
* 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:
#include "main/errors.h"
-#include "main/imports.h"
+
#include "main/hash.h"
#include "main/mtypes.h"
#include "program/prog_parameter.h"
#include "program/programopt.h"
#include "compiler/nir/nir.h"
+#include "compiler/nir/nir_serialize.h"
#include "draw/draw_context.h"
#include "pipe/p_context.h"
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_ureg.h"
+#include "util/u_memory.h"
+
#include "st_debug.h"
#include "st_cb_bitmap.h"
#include "st_cb_drawpixels.h"
#include "cso_cache/cso_context.h"
+static void
+destroy_program_variants(struct st_context *st, struct gl_program *target);
static void
set_affected_state_flags(uint64_t *states,
*/
}
+/**
+ * Free all basic program variants and unref program.
+ */
+void
+st_release_program(struct st_context *st, struct st_program **p)
+{
+ if (!*p)
+ return;
+
+ destroy_program_variants(st, &((*p)->Base));
+ st_reference_prog(st, p, NULL);
+}
+
void
st_finalize_nir_before_variants(struct nir_shader *nir)
{
NIR_PASS_V(nir, st_nir_lower_wpos_ytransform, prog, screen);
NIR_PASS_V(nir, nir_lower_system_values);
+ NIR_PASS_V(nir, nir_lower_compute_system_values, NULL);
/* Optimise NIR */
NIR_PASS_V(nir, nir_opt_constant_folding);
if (stp->Base.arb.IsPositionInvariant)
_mesa_insert_mvp_code(st->ctx, &stp->Base);
- st_prepare_vertex_program(stp);
-
/* ARB_vp: */
if (!stp->glsl_to_tgsi) {
_mesa_remove_output_reads(&stp->Base, PROGRAM_OUTPUT);
stp->affected_states |= ST_NEW_VS_CONSTANTS;
/* Translate to NIR if preferred. */
- if (st->pipe->screen->get_shader_param(st->pipe->screen,
+ if (PIPE_SHADER_IR_NIR ==
+ st->pipe->screen->get_shader_param(st->pipe->screen,
PIPE_SHADER_VERTEX,
PIPE_SHADER_CAP_PREFERRED_IR)) {
assert(!stp->glsl_to_tgsi);
if (stp->Base.nir)
ralloc_free(stp->Base.nir);
+ if (stp->serialized_nir) {
+ free(stp->serialized_nir);
+ stp->serialized_nir = NULL;
+ }
+
stp->state.type = PIPE_SHADER_IR_NIR;
stp->Base.nir = st_translate_prog_to_nir(st, &stp->Base,
MESA_SHADER_VERTEX);
+
+ /* We must update stp->Base.info after translation and before
+ * st_prepare_vertex_program is called, because inputs_read
+ * may become outdated after NIR optimization passes.
+ *
+ * For ffvp/ARB_vp inputs_read is populated based
+ * on declared attributes without taking their usage into
+ * consideration. When creating shader variants we expect
+ * that their inputs_read would match the base ones for
+ * input mapping to work properly.
+ */
+ nir_shader_gather_info(stp->Base.nir,
+ nir_shader_get_entrypoint(stp->Base.nir));
+ st_nir_assign_vs_in_locations(stp->Base.nir);
+ stp->Base.info = stp->Base.nir->info;
+
/* For st_draw_feedback, we need to generate TGSI too if draw doesn't
* use LLVM.
*/
- if (draw_has_llvm())
+ /* TODO: Draw can't handle lowered IO. */
+ if (draw_has_llvm() && !stp->Base.info.io_lowered) {
+ st_prepare_vertex_program(stp);
return true;
+ }
}
}
+ st_prepare_vertex_program(stp);
+
/* Get semantic names and indices. */
for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
if (stp->Base.info.outputs_written & BITFIELD64_BIT(attr)) {
if (ureg == NULL)
return false;
- if (stp->Base.info.clip_distance_array_size)
- ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
- stp->Base.info.clip_distance_array_size);
- if (stp->Base.info.cull_distance_array_size)
- ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
- stp->Base.info.cull_distance_array_size);
+ ureg_setup_shader_info(ureg, &stp->Base.info);
if (ST_DEBUG & DEBUG_MESA) {
_mesa_print_program(&stp->Base);
return stp->state.tokens != NULL;
}
+static struct nir_shader *
+get_nir_shader(struct st_context *st, struct st_program *stp)
+{
+ if (stp->Base.nir) {
+ nir_shader *nir = stp->Base.nir;
+
+ /* The first shader variant takes ownership of NIR, so that there is
+ * no cloning. Additional shader variants are always generated from
+ * serialized NIR to save memory.
+ */
+ stp->Base.nir = NULL;
+ assert(stp->serialized_nir && stp->serialized_nir_size);
+ return nir;
+ }
+
+ struct blob_reader blob_reader;
+ const struct nir_shader_compiler_options *options =
+ st->ctx->Const.ShaderCompilerOptions[stp->Base.info.stage].NirOptions;
+
+ blob_reader_init(&blob_reader, stp->serialized_nir, stp->serialized_nir_size);
+ return nir_deserialize(NULL, options, &blob_reader);
+}
+
+static void
+lower_ucp(struct st_context *st,
+ struct nir_shader *nir,
+ unsigned ucp_enables,
+ struct gl_program_parameter_list *params)
+{
+ if (nir->info.outputs_written & VARYING_BIT_CLIP_DIST0)
+ NIR_PASS_V(nir, nir_lower_clip_disable, ucp_enables);
+ else {
+ struct pipe_screen *screen = st->pipe->screen;
+ bool can_compact = screen->get_param(screen,
+ PIPE_CAP_NIR_COMPACT_ARRAYS);
+ bool use_eye = st->ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX] != NULL;
+
+ gl_state_index16 clipplane_state[MAX_CLIP_PLANES][STATE_LENGTH];
+ for (int i = 0; i < MAX_CLIP_PLANES; ++i) {
+ if (use_eye) {
+ clipplane_state[i][0] = STATE_CLIPPLANE;
+ clipplane_state[i][1] = i;
+ } else {
+ clipplane_state[i][0] = STATE_INTERNAL;
+ clipplane_state[i][1] = STATE_CLIP_INTERNAL;
+ clipplane_state[i][2] = i;
+ }
+ _mesa_add_state_reference(params, clipplane_state[i]);
+ }
+
+ if (nir->info.stage == MESA_SHADER_VERTEX) {
+ NIR_PASS_V(nir, nir_lower_clip_vs, ucp_enables,
+ true, can_compact, clipplane_state);
+ } else if (nir->info.stage == MESA_SHADER_GEOMETRY) {
+ NIR_PASS_V(nir, nir_lower_clip_gs, ucp_enables,
+ can_compact, clipplane_state);
+ }
+
+ NIR_PASS_V(nir, nir_lower_io_to_temporaries,
+ nir_shader_get_entrypoint(nir), true, false);
+ NIR_PASS_V(nir, nir_lower_global_vars_to_local);
+ }
+}
+
static const gl_state_index16 depth_range_state[STATE_LENGTH] =
{ STATE_DEPTH_RANGE };
{
struct st_common_variant *vpv = CALLOC_STRUCT(st_common_variant);
struct pipe_context *pipe = st->pipe;
- struct pipe_screen *screen = pipe->screen;
struct pipe_shader_state state = {0};
static const gl_state_index16 point_size_state[STATE_LENGTH] =
state.stream_output = stvp->state.stream_output;
if (stvp->state.type == PIPE_SHADER_IR_NIR &&
- (!key->is_draw_shader || draw_has_llvm())) {
+ (!key->is_draw_shader ||
+ /* TODO: Draw can't handle lowered IO. */
+ (draw_has_llvm() && !stvp->Base.info.io_lowered))) {
bool finalize = false;
state.type = PIPE_SHADER_IR_NIR;
- state.ir.nir = nir_shader_clone(NULL, stvp->Base.nir);
+ state.ir.nir = get_nir_shader(st, stvp);
if (key->clamp_color) {
NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
finalize = true;
}
if (key->lower_ucp) {
- bool can_compact = screen->get_param(screen,
- PIPE_CAP_NIR_COMPACT_ARRAYS);
-
- bool use_eye = st->ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX] != NULL;
- gl_state_index16 clipplane_state[MAX_CLIP_PLANES][STATE_LENGTH];
- for (int i = 0; i < MAX_CLIP_PLANES; ++i) {
- if (use_eye) {
- clipplane_state[i][0] = STATE_CLIPPLANE;
- clipplane_state[i][1] = i;
- } else {
- clipplane_state[i][0] = STATE_INTERNAL;
- clipplane_state[i][1] = STATE_CLIP_INTERNAL;
- clipplane_state[i][2] = i;
- }
- _mesa_add_state_reference(params, clipplane_state[i]);
- }
-
- NIR_PASS_V(state.ir.nir, nir_lower_clip_vs, key->lower_ucp,
- true, can_compact, clipplane_state);
- NIR_PASS_V(state.ir.nir, nir_lower_io_to_temporaries,
- nir_shader_get_entrypoint(state.ir.nir), true, false);
- NIR_PASS_V(state.ir.nir, nir_lower_global_vars_to_local);
+ lower_ucp(st, state.ir.nir, key->lower_ucp, params);
finalize = true;
}
/* Translate to NIR. */
if (!stfp->ati_fs &&
+ PIPE_SHADER_IR_NIR ==
st->pipe->screen->get_shader_param(st->pipe->screen,
PIPE_SHADER_FRAGMENT,
PIPE_SHADER_CAP_PREFERRED_IR)) {
if (stfp->Base.nir)
ralloc_free(stfp->Base.nir);
+ if (stfp->serialized_nir) {
+ free(stfp->serialized_nir);
+ stfp->serialized_nir = NULL;
+ }
stfp->state.type = PIPE_SHADER_IR_NIR;
stfp->Base.nir = nir;
return true;
if (ureg == NULL)
return false;
+ ureg_setup_shader_info(ureg, &stfp->Base.info);
+
if (ST_DEBUG & DEBUG_MESA) {
_mesa_print_program(&stfp->Base);
_mesa_print_program_parameters(st->ctx, &stfp->Base);
if (write_all == GL_TRUE)
ureg_property(ureg, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS, 1);
- if (stfp->Base.info.fs.depth_layout != FRAG_DEPTH_LAYOUT_NONE) {
- switch (stfp->Base.info.fs.depth_layout) {
- case FRAG_DEPTH_LAYOUT_ANY:
- ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
- TGSI_FS_DEPTH_LAYOUT_ANY);
- break;
- case FRAG_DEPTH_LAYOUT_GREATER:
- ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
- TGSI_FS_DEPTH_LAYOUT_GREATER);
- break;
- case FRAG_DEPTH_LAYOUT_LESS:
- ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
- TGSI_FS_DEPTH_LAYOUT_LESS);
- break;
- case FRAG_DEPTH_LAYOUT_UNCHANGED:
- ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
- TGSI_FS_DEPTH_LAYOUT_UNCHANGED);
- break;
- default:
- assert(0);
- }
- }
-
if (stfp->glsl_to_tgsi) {
st_translate_program(st->ctx,
PIPE_SHADER_FRAGMENT,
bool finalize = false;
state.type = PIPE_SHADER_IR_NIR;
- state.ir.nir = nir_shader_clone(NULL, stfp->Base.nir);
+ state.ir.nir = get_nir_shader(st, stfp);
if (key->clamp_color) {
NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
}
if (key->lower_two_sided_color) {
- NIR_PASS_V(state.ir.nir, nir_lower_two_sided_color);
+ bool face_sysval = st->ctx->Const.GLSLFrontFacingIsSysVal;
+ NIR_PASS_V(state.ir.nir, nir_lower_two_sided_color, face_sysval);
finalize = true;
}
if (key->persample_shading) {
nir_shader *shader = state.ir.nir;
- nir_foreach_variable(var, &shader->inputs)
+ nir_foreach_shader_in_variable(var, shader)
var->data.sample = true;
finalize = true;
}
if (ureg == NULL)
return false;
- switch (stage) {
- case PIPE_SHADER_TESS_CTRL:
- ureg_property(ureg, TGSI_PROPERTY_TCS_VERTICES_OUT,
- stp->Base.info.tess.tcs_vertices_out);
- break;
-
- case PIPE_SHADER_TESS_EVAL:
- if (stp->Base.info.tess.primitive_mode == GL_ISOLINES)
- ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE, GL_LINES);
- else
- ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE,
- stp->Base.info.tess.primitive_mode);
-
- STATIC_ASSERT((TESS_SPACING_EQUAL + 1) % 3 == PIPE_TESS_SPACING_EQUAL);
- STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD + 1) % 3 ==
- PIPE_TESS_SPACING_FRACTIONAL_ODD);
- STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN + 1) % 3 ==
- PIPE_TESS_SPACING_FRACTIONAL_EVEN);
-
- ureg_property(ureg, TGSI_PROPERTY_TES_SPACING,
- (stp->Base.info.tess.spacing + 1) % 3);
-
- ureg_property(ureg, TGSI_PROPERTY_TES_VERTEX_ORDER_CW,
- !stp->Base.info.tess.ccw);
- ureg_property(ureg, TGSI_PROPERTY_TES_POINT_MODE,
- stp->Base.info.tess.point_mode);
- break;
-
- case PIPE_SHADER_GEOMETRY:
- ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM,
- stp->Base.info.gs.input_primitive);
- ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM,
- stp->Base.info.gs.output_primitive);
- ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES,
- stp->Base.info.gs.vertices_out);
- ureg_property(ureg, TGSI_PROPERTY_GS_INVOCATIONS,
- stp->Base.info.gs.invocations);
- break;
-
- default:
- break;
- }
+ ureg_setup_shader_info(ureg, &stp->Base.info);
ubyte inputSlotToAttr[VARYING_SLOT_TESS_MAX];
ubyte inputMapping[VARYING_SLOT_TESS_MAX];
memset(outputMapping, 0, sizeof(outputMapping));
memset(&stp->state, 0, sizeof(stp->state));
- if (prog->info.clip_distance_array_size)
- ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
- prog->info.clip_distance_array_size);
- if (prog->info.cull_distance_array_size)
- ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
- prog->info.cull_distance_array_size);
-
/*
* Convert Mesa program inputs to TGSI input register semantics.
*/
struct pipe_context *pipe = st->pipe;
struct st_variant *v;
struct pipe_shader_state state = {0};
+ struct gl_program_parameter_list *params = prog->Base.Parameters;
/* Search for existing variant */
for (v = prog->variants; v; v = v->next) {
bool finalize = false;
state.type = PIPE_SHADER_IR_NIR;
- state.ir.nir = nir_shader_clone(NULL, prog->Base.nir);
+ state.ir.nir = get_nir_shader(st, prog);
if (key->clamp_color) {
NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
finalize = true;
}
+ if (key->lower_ucp) {
+ lower_ucp(st, state.ir.nir, key->lower_ucp, params);
+ finalize = true;
+ }
+
state.stream_output = prog->state.stream_output;
if (finalize || !st->allow_st_finalize_nir_twice) {
}
}
+void
+st_serialize_nir(struct st_program *stp)
+{
+ if (!stp->serialized_nir) {
+ struct blob blob;
+ size_t size;
+
+ blob_init(&blob);
+ nir_serialize(&blob, stp->Base.nir, false);
+ blob_finish_get_buffer(&blob, &stp->serialized_nir, &size);
+ stp->serialized_nir_size = size;
+ }
+}
+
void
st_finalize_program(struct st_context *st, struct gl_program *prog)
{
st->dirty |= ((struct st_program *)prog)->affected_states;
}
- if (prog->nir)
+ if (prog->nir) {
nir_sweep(prog->nir);
+ /* This is only needed for ARB_vp/fp programs and when the disk cache
+ * is disabled. If the disk cache is enabled, GLSL programs are
+ * serialized in write_nir_to_cache.
+ */
+ st_serialize_nir(st_program(prog));
+ }
+
/* Create Gallium shaders now instead of on demand. */
if (ST_DEBUG & DEBUG_PRECOMPILE ||
st->shader_has_one_variant[prog->info.stage])
projects / mesa.git / blobdiff