*/
+#include "main/errors.h"
#include "main/imports.h"
#include "main/hash.h"
#include "main/mtypes.h"
#include "program/prog_parameter.h"
#include "program/prog_print.h"
+#include "program/prog_to_nir.h"
#include "program/programopt.h"
#include "compiler/nir/nir.h"
#include "st_mesa_to_tgsi.h"
#include "st_atifs_to_tgsi.h"
#include "st_nir.h"
+#include "st_shader_cache.h"
#include "cso_cache/cso_context.h"
set_affected_state_flags(states, prog,
ST_NEW_VS_CONSTANTS,
ST_NEW_VS_SAMPLER_VIEWS,
- ST_NEW_RENDER_SAMPLERS,
+ ST_NEW_VS_SAMPLERS,
ST_NEW_VS_IMAGES,
ST_NEW_VS_UBOS,
ST_NEW_VS_SSBOS,
break;
case MESA_SHADER_TESS_CTRL:
- states = &((struct st_tessctrl_program*)prog)->affected_states;
+ states = &(st_common_program(prog))->affected_states;
*states = ST_NEW_TCS_STATE;
set_affected_state_flags(states, prog,
ST_NEW_TCS_CONSTANTS,
ST_NEW_TCS_SAMPLER_VIEWS,
- ST_NEW_RENDER_SAMPLERS,
+ ST_NEW_TCS_SAMPLERS,
ST_NEW_TCS_IMAGES,
ST_NEW_TCS_UBOS,
ST_NEW_TCS_SSBOS,
break;
case MESA_SHADER_TESS_EVAL:
- states = &((struct st_tesseval_program*)prog)->affected_states;
+ states = &(st_common_program(prog))->affected_states;
*states = ST_NEW_TES_STATE |
ST_NEW_RASTERIZER;
set_affected_state_flags(states, prog,
ST_NEW_TES_CONSTANTS,
ST_NEW_TES_SAMPLER_VIEWS,
- ST_NEW_RENDER_SAMPLERS,
+ ST_NEW_TES_SAMPLERS,
ST_NEW_TES_IMAGES,
ST_NEW_TES_UBOS,
ST_NEW_TES_SSBOS,
break;
case MESA_SHADER_GEOMETRY:
- states = &((struct st_geometry_program*)prog)->affected_states;
+ states = &(st_common_program(prog))->affected_states;
*states = ST_NEW_GS_STATE |
ST_NEW_RASTERIZER;
set_affected_state_flags(states, prog,
ST_NEW_GS_CONSTANTS,
ST_NEW_GS_SAMPLER_VIEWS,
- ST_NEW_RENDER_SAMPLERS,
+ ST_NEW_GS_SAMPLERS,
ST_NEW_GS_IMAGES,
ST_NEW_GS_UBOS,
ST_NEW_GS_SSBOS,
set_affected_state_flags(states, prog,
ST_NEW_FS_CONSTANTS,
ST_NEW_FS_SAMPLER_VIEWS,
- ST_NEW_RENDER_SAMPLERS,
+ ST_NEW_FS_SAMPLERS,
ST_NEW_FS_IMAGES,
ST_NEW_FS_UBOS,
ST_NEW_FS_SSBOS,
}
}
+static void
+delete_ir(struct pipe_shader_state *ir)
+{
+ if (ir->tokens)
+ ureg_free_tokens(ir->tokens);
+
+ /* Note: Any setup of ->ir.nir that has had pipe->create_*_state called on
+ * it has resulted in the driver taking ownership of the NIR. Those
+ * callers should be NULLing out the nir field in any pipe_shader_state
+ * that might have this called in order to indicate that.
+ *
+ * GLSL IR and ARB programs will have set gl_program->nir to the same
+ * shader as ir->ir.nir, so it will be freed by _mesa_delete_program().
+ */
+}
+
/**
* Delete a vertex program variant. Note the caller must unlink
* the variant from the linked list.
if (vpv->draw_shader)
draw_delete_vertex_shader( st->draw, vpv->draw_shader );
- if (((vpv->tgsi.type == PIPE_SHADER_IR_TGSI)) && vpv->tgsi.tokens)
- ureg_free_tokens(vpv->tgsi.tokens);
+ delete_ir(&vpv->tgsi);
free( vpv );
}
stvp->variants = NULL;
- if ((stvp->tgsi.type == PIPE_SHADER_IR_TGSI) && stvp->tgsi.tokens) {
- tgsi_free_tokens(stvp->tgsi.tokens);
- stvp->tgsi.tokens = NULL;
- }
+ delete_ir(&stvp->tgsi);
}
stfp->variants = NULL;
- if ((stfp->tgsi.type == PIPE_SHADER_IR_TGSI) && stfp->tgsi.tokens) {
- ureg_free_tokens(stfp->tgsi.tokens);
- stfp->tgsi.tokens = NULL;
- }
+ delete_ir(&stfp->tgsi);
}
*variants = NULL;
- if (tgsi->tokens) {
- ureg_free_tokens(tgsi->tokens);
- tgsi->tokens = NULL;
- }
+ delete_ir(tgsi);
}
*variants = NULL;
if (stcp->tgsi.prog) {
- ureg_free_tokens(stcp->tgsi.prog);
- stcp->tgsi.prog = NULL;
+ switch (stcp->tgsi.ir_type) {
+ case PIPE_SHADER_IR_TGSI:
+ ureg_free_tokens(stcp->tgsi.prog);
+ stcp->tgsi.prog = NULL;
+ break;
+ case PIPE_SHADER_IR_NIR:
+ /* pipe driver took ownership of prog */
+ break;
+ case PIPE_SHADER_IR_NATIVE:
+ /* ??? */
+ stcp->tgsi.prog = NULL;
+ break;
+ }
}
}
+/**
+ * Translate ARB (asm) program to NIR
+ */
+static nir_shader *
+st_translate_prog_to_nir(struct st_context *st, struct gl_program *prog,
+ gl_shader_stage stage)
+{
+ enum pipe_shader_type p_stage = stage; /* valid for VS/FS */
+ const bool is_scalar =
+ st->pipe->screen->get_shader_param(st->pipe->screen, p_stage,
+ PIPE_SHADER_CAP_SCALAR_ISA);
+
+ const struct gl_shader_compiler_options *options =
+ &st->ctx->Const.ShaderCompilerOptions[stage];
+
+ /* Translate to NIR */
+ nir_shader *nir = prog_to_nir(prog, options->NirOptions);
+ NIR_PASS_V(nir, nir_lower_regs_to_ssa); /* turn registers into SSA */
+ nir_validate_shader(nir, "after st/ptn lower_regs_to_ssa");
+
+ NIR_PASS_V(nir, st_nir_lower_wpos_ytransform, prog, st->pipe->screen);
+ NIR_PASS_V(nir, nir_lower_system_values);
+
+ /* Optimise NIR */
+ NIR_PASS_V(nir, nir_opt_constant_folding);
+ st_nir_opts(nir, is_scalar);
+ nir_validate_shader(nir, "after st/ptn NIR opts");
+
+ return nir;
+}
/**
* Translate a vertex program.
enum pipe_error error;
unsigned num_outputs = 0;
unsigned attr;
- unsigned input_to_index[VERT_ATTRIB_MAX] = {0};
- unsigned output_slot_to_attr[VARYING_SLOT_MAX] = {0};
ubyte output_semantic_name[VARYING_SLOT_MAX] = {0};
ubyte output_semantic_index[VARYING_SLOT_MAX] = {0};
stvp->num_inputs = 0;
+ memset(stvp->input_to_index, ~0, sizeof(stvp->input_to_index));
if (stvp->Base.arb.IsPositionInvariant)
_mesa_insert_mvp_code(st->ctx, &stvp->Base);
*/
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
if ((stvp->Base.info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
- input_to_index[attr] = stvp->num_inputs;
+ stvp->input_to_index[attr] = stvp->num_inputs;
stvp->index_to_input[stvp->num_inputs] = attr;
stvp->num_inputs++;
- if ((stvp->Base.info.double_inputs_read &
- BITFIELD64_BIT(attr)) != 0) {
+ if ((stvp->Base.DualSlotInputs & BITFIELD64_BIT(attr)) != 0) {
/* add placeholder for second part of a double attribute */
stvp->index_to_input[stvp->num_inputs] = ST_DOUBLE_ATTRIB_PLACEHOLDER;
stvp->num_inputs++;
}
}
/* bit of a hack, presetup potentially unused edgeflag input */
- input_to_index[VERT_ATTRIB_EDGEFLAG] = stvp->num_inputs;
+ stvp->input_to_index[VERT_ATTRIB_EDGEFLAG] = stvp->num_inputs;
stvp->index_to_input[stvp->num_inputs] = VERT_ATTRIB_EDGEFLAG;
/* Compute mapping of vertex program outputs to slots.
unsigned slot = num_outputs++;
stvp->result_to_output[attr] = slot;
- output_slot_to_attr[slot] = attr;
- switch (attr) {
- case VARYING_SLOT_POS:
- output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_COL0:
- output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_COL1:
- output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- output_semantic_index[slot] = 1;
- break;
- case VARYING_SLOT_BFC0:
- output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_BFC1:
- output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
- output_semantic_index[slot] = 1;
- break;
- case VARYING_SLOT_FOGC:
- output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_PSIZ:
- output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_CLIP_DIST0:
- output_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_CLIP_DIST1:
- output_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
- output_semantic_index[slot] = 1;
- break;
- case VARYING_SLOT_CULL_DIST0:
- case VARYING_SLOT_CULL_DIST1:
- /* these should have been lowered by GLSL */
- assert(0);
- break;
- case VARYING_SLOT_EDGE:
- assert(0);
- break;
- case VARYING_SLOT_CLIP_VERTEX:
- output_semantic_name[slot] = TGSI_SEMANTIC_CLIPVERTEX;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_LAYER:
- output_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_VIEWPORT:
- output_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
- output_semantic_index[slot] = 0;
- break;
-
- case VARYING_SLOT_TEX0:
- case VARYING_SLOT_TEX1:
- case VARYING_SLOT_TEX2:
- case VARYING_SLOT_TEX3:
- case VARYING_SLOT_TEX4:
- case VARYING_SLOT_TEX5:
- case VARYING_SLOT_TEX6:
- case VARYING_SLOT_TEX7:
- if (st->needs_texcoord_semantic) {
- output_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
- output_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
- break;
- }
- /* fall through */
- case VARYING_SLOT_VAR0:
- default:
- assert(attr >= VARYING_SLOT_VAR0 ||
- (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
- output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- output_semantic_index[slot] =
- st_get_generic_varying_index(st, attr);
- break;
- }
+ unsigned semantic_name, semantic_index;
+ tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
+ &semantic_name, &semantic_index);
+ output_semantic_name[slot] = semantic_name;
+ output_semantic_index[slot] = semantic_index;
}
}
/* similar hack to above, presetup potentially unused edgeflag output */
}
if (stvp->shader_program) {
- nir_shader *nir = st_glsl_to_nir(st, &stvp->Base, stvp->shader_program,
- MESA_SHADER_VERTEX);
-
- stvp->tgsi.type = PIPE_SHADER_IR_NIR;
- stvp->tgsi.ir.nir = nir;
-
- struct gl_program *prog = stvp->shader_program->last_vert_prog;
- if (prog) {
- st_translate_stream_output_info2(prog->sh.LinkedTransformFeedback,
- stvp->result_to_output,
- &stvp->tgsi.stream_output);
- }
+ st_translate_stream_output_info(stvp->Base.sh.LinkedTransformFeedback,
+ stvp->result_to_output,
+ &stvp->tgsi.stream_output);
+ st_store_ir_in_disk_cache(st, &stvp->Base, true);
return true;
}
&stvp->Base,
/* inputs */
stvp->num_inputs,
- input_to_index,
+ stvp->input_to_index,
NULL, /* inputSlotToAttr */
NULL, /* input semantic name */
NULL, /* input semantic index */
/* outputs */
num_outputs,
stvp->result_to_output,
- output_slot_to_attr,
output_semantic_name,
output_semantic_index);
- st_translate_stream_output_info(stvp->glsl_to_tgsi,
+ st_translate_stream_output_info(stvp->Base.sh.LinkedTransformFeedback,
stvp->result_to_output,
&stvp->tgsi.stream_output);
free_glsl_to_tgsi_visitor(stvp->glsl_to_tgsi);
- stvp->glsl_to_tgsi = NULL;
} else
error = st_translate_mesa_program(st->ctx,
PIPE_SHADER_VERTEX,
&stvp->Base,
/* inputs */
stvp->num_inputs,
- input_to_index,
+ stvp->input_to_index,
NULL, /* input semantic name */
NULL, /* input semantic index */
NULL,
return false;
}
- stvp->tgsi.tokens = ureg_get_tokens(ureg, NULL);
+ stvp->tgsi.tokens = ureg_get_tokens(ureg, &stvp->num_tgsi_tokens);
ureg_destroy(ureg);
+
+ if (stvp->glsl_to_tgsi) {
+ stvp->glsl_to_tgsi = NULL;
+ st_store_ir_in_disk_cache(st, &stvp->Base, false);
+ }
+
+ bool use_nir = PIPE_SHADER_IR_NIR ==
+ st->pipe->screen->get_shader_param(st->pipe->screen, PIPE_SHADER_VERTEX,
+ PIPE_SHADER_CAP_PREFERRED_IR);
+
+ if (use_nir) {
+ nir_shader *nir =
+ st_translate_prog_to_nir(st, &stvp->Base, MESA_SHADER_VERTEX);
+
+ if (stvp->tgsi.ir.nir)
+ ralloc_free(stvp->tgsi.ir.nir);
+ stvp->tgsi.type = PIPE_SHADER_IR_NIR;
+ stvp->tgsi.ir.nir = nir;
+ stvp->Base.nir = nir;
+ return true;
+ }
+
return stvp->tgsi.tokens != NULL;
}
vpv->tgsi.stream_output = stvp->tgsi.stream_output;
vpv->num_inputs = stvp->num_inputs;
+ /* When generating a NIR program, we usually don't have TGSI tokens.
+ * However, we do create them for ARB_vertex_program / fixed-function VS
+ * programs which we may need to use with the draw module for legacy
+ * feedback/select emulation. If they exist, copy them.
+ */
+ if (stvp->tgsi.tokens)
+ vpv->tgsi.tokens = tgsi_dup_tokens(stvp->tgsi.tokens);
+
if (stvp->tgsi.type == PIPE_SHADER_IR_NIR) {
vpv->tgsi.type = PIPE_SHADER_IR_NIR;
vpv->tgsi.ir.nir = nir_shader_clone(NULL, stvp->tgsi.ir.nir);
if (key->clamp_color)
NIR_PASS_V(vpv->tgsi.ir.nir, nir_lower_clamp_color_outputs);
- if (key->passthrough_edgeflags)
+ if (key->passthrough_edgeflags) {
NIR_PASS_V(vpv->tgsi.ir.nir, nir_lower_passthrough_edgeflags);
+ vpv->num_inputs++;
+ }
- st_finalize_nir(st, &stvp->Base, vpv->tgsi.ir.nir);
+ st_finalize_nir(st, &stvp->Base, stvp->shader_program,
+ vpv->tgsi.ir.nir);
vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->tgsi);
/* driver takes ownership of IR: */
return vpv;
}
- vpv->tgsi.tokens = tgsi_dup_tokens(stvp->tgsi.tokens);
-
/* Emulate features. */
if (key->clamp_color || key->passthrough_edgeflags) {
const struct tgsi_token *tokens;
/* create now */
vpv = st_create_vp_variant(st, stvp, key);
if (vpv) {
+ for (unsigned index = 0; index < vpv->num_inputs; ++index) {
+ unsigned attr = stvp->index_to_input[index];
+ if (attr == ST_DOUBLE_ATTRIB_PLACEHOLDER)
+ continue;
+ vpv->vert_attrib_mask |= 1u << attr;
+ }
+
/* insert into list */
vpv->next = stvp->variants;
stvp->variants = vpv;
st_translate_fragment_program(struct st_context *st,
struct st_fragment_program *stfp)
{
- GLuint outputMapping[2 * FRAG_RESULT_MAX];
- GLuint inputMapping[VARYING_SLOT_MAX];
- GLuint inputSlotToAttr[VARYING_SLOT_MAX];
- GLuint interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */
+ /* We have already compiled to NIR so just return */
+ if (stfp->shader_program) {
+ st_store_ir_in_disk_cache(st, &stfp->Base, true);
+ return true;
+ }
+
+ ubyte outputMapping[2 * FRAG_RESULT_MAX];
+ ubyte inputMapping[VARYING_SLOT_MAX];
+ ubyte inputSlotToAttr[VARYING_SLOT_MAX];
+ ubyte interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */
GLuint attr;
GLbitfield64 inputsRead;
struct ureg_program *ureg;
if (stfp->ati_fs) {
/* Just set them for ATI_fs unconditionally. */
stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
- ST_NEW_RENDER_SAMPLERS;
+ ST_NEW_FS_SAMPLERS;
} else {
/* ARB_fp */
if (stfp->Base.SamplersUsed)
stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
- ST_NEW_RENDER_SAMPLERS;
+ ST_NEW_FS_SAMPLERS;
}
}
+
+ bool use_nir = PIPE_SHADER_IR_NIR ==
+ st->pipe->screen->get_shader_param(st->pipe->screen,
+ PIPE_SHADER_FRAGMENT,
+ PIPE_SHADER_CAP_PREFERRED_IR);
+
+ if (use_nir && !stfp->ati_fs) {
+ nir_shader *nir =
+ st_translate_prog_to_nir(st, &stfp->Base, MESA_SHADER_FRAGMENT);
+
+ if (stfp->tgsi.ir.nir)
+ ralloc_free(stfp->tgsi.ir.nir);
+ stfp->tgsi.type = PIPE_SHADER_IR_NIR;
+ stfp->tgsi.ir.nir = nir;
+ stfp->Base.nir = nir;
+ return true;
+ }
+
/*
* Convert Mesa program inputs to TGSI input register semantics.
*/
}
}
- if (stfp->shader_program) {
- nir_shader *nir = st_glsl_to_nir(st, &stfp->Base, stfp->shader_program,
- MESA_SHADER_FRAGMENT);
-
- stfp->tgsi.type = PIPE_SHADER_IR_NIR;
- stfp->tgsi.ir.nir = nir;
-
- return true;
- }
-
ureg = ureg_create_with_screen(PIPE_SHADER_FRAGMENT, st->pipe->screen);
if (ureg == NULL)
return false;
/* outputs */
fs_num_outputs,
outputMapping,
- NULL,
fs_output_semantic_name,
fs_output_semantic_index);
free_glsl_to_tgsi_visitor(stfp->glsl_to_tgsi);
- stfp->glsl_to_tgsi = NULL;
} else if (stfp->ati_fs)
st_translate_atifs_program(ureg,
stfp->ati_fs,
fs_output_semantic_name,
fs_output_semantic_index);
- stfp->tgsi.tokens = ureg_get_tokens(ureg, NULL);
+ stfp->tgsi.tokens = ureg_get_tokens(ureg, &stfp->num_tgsi_tokens);
ureg_destroy(ureg);
+
+ if (stfp->glsl_to_tgsi) {
+ stfp->glsl_to_tgsi = NULL;
+ st_store_ir_in_disk_cache(st, &stfp->Base, false);
+ }
+
return stfp->tgsi.tokens != NULL;
}
struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant);
struct pipe_shader_state tgsi = {0};
struct gl_program_parameter_list *params = stfp->Base.Parameters;
- static const gl_state_index texcoord_state[STATE_LENGTH] =
+ static const gl_state_index16 texcoord_state[STATE_LENGTH] =
{ STATE_INTERNAL, STATE_CURRENT_ATTRIB, VERT_ATTRIB_TEX0 };
- static const gl_state_index scale_state[STATE_LENGTH] =
+ static const gl_state_index16 scale_state[STATE_LENGTH] =
{ STATE_INTERNAL, STATE_PT_SCALE };
- static const gl_state_index bias_state[STATE_LENGTH] =
+ static const gl_state_index16 bias_state[STATE_LENGTH] =
{ STATE_INTERNAL, STATE_PT_BIAS };
if (!variant)
NIR_PASS_V(tgsi.ir.nir, nir_lower_tex, &options);
}
- st_finalize_nir(st, &stfp->Base, tgsi.ir.nir);
+ st_finalize_nir(st, &stfp->Base, stfp->shader_program, tgsi.ir.nir);
if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv)) {
/* This pass needs to happen *after* nir_lower_sampler */
key->external.lower_iyuv);
}
+ /* Some of the lowering above may have introduced new varyings */
+ nir_shader_gather_info(tgsi.ir.nir,
+ nir_shader_get_entrypoint(tgsi.ir.nir));
+
variant->driver_shader = pipe->create_fs_state(pipe, &tgsi);
variant->key = *key;
/* create new */
fpv = st_create_fp_variant(st, stfp, key);
if (fpv) {
- /* insert into list */
- fpv->next = stfp->variants;
- stfp->variants = fpv;
+ if (key->bitmap || key->drawpixels) {
+ /* Regular variants should always come before the
+ * bitmap & drawpixels variants, (unless there
+ * are no regular variants) so that
+ * st_update_fp can take a fast path when
+ * shader_has_one_variant is set.
+ */
+ if (!stfp->variants) {
+ stfp->variants = fpv;
+ } else {
+ /* insert into list after the first one */
+ fpv->next = stfp->variants->next;
+ stfp->variants->next = fpv;
+ }
+ } else {
+ /* insert into list */
+ fpv->next = stfp->variants;
+ stfp->variants = fpv;
+ }
}
}
unsigned tgsi_processor,
struct pipe_shader_state *out_state)
{
- GLuint inputSlotToAttr[VARYING_SLOT_TESS_MAX];
- GLuint inputMapping[VARYING_SLOT_TESS_MAX];
- GLuint outputSlotToAttr[VARYING_SLOT_TESS_MAX];
- GLuint outputMapping[VARYING_SLOT_TESS_MAX];
+ ubyte inputSlotToAttr[VARYING_SLOT_TESS_MAX];
+ ubyte inputMapping[VARYING_SLOT_TESS_MAX];
+ ubyte outputMapping[VARYING_SLOT_TESS_MAX];
GLuint attr;
ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
memset(inputSlotToAttr, 0, sizeof(inputSlotToAttr));
memset(inputMapping, 0, sizeof(inputMapping));
- memset(outputSlotToAttr, 0, sizeof(outputSlotToAttr));
memset(outputMapping, 0, sizeof(outputMapping));
memset(out_state, 0, sizeof(*out_state));
* Convert Mesa program inputs to TGSI input register semantics.
*/
for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
- if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
- const GLuint slot = num_inputs++;
+ if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) == 0)
+ continue;
- inputMapping[attr] = slot;
- inputSlotToAttr[slot] = attr;
+ unsigned slot = num_inputs++;
- switch (attr) {
- case VARYING_SLOT_PRIMITIVE_ID:
- assert(tgsi_processor == PIPE_SHADER_GEOMETRY);
- input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
- input_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_POS:
- input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
- input_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_COL0:
- input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- input_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_COL1:
- input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- input_semantic_index[slot] = 1;
- break;
- case VARYING_SLOT_FOGC:
- input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
- input_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_CLIP_VERTEX:
- input_semantic_name[slot] = TGSI_SEMANTIC_CLIPVERTEX;
- input_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_CLIP_DIST0:
- input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
- input_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_CLIP_DIST1:
- input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
- input_semantic_index[slot] = 1;
- break;
- case VARYING_SLOT_CULL_DIST0:
- case VARYING_SLOT_CULL_DIST1:
- /* these should have been lowered by GLSL */
- assert(0);
- break;
- case VARYING_SLOT_PSIZ:
- input_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
- input_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_TEX0:
- case VARYING_SLOT_TEX1:
- case VARYING_SLOT_TEX2:
- case VARYING_SLOT_TEX3:
- case VARYING_SLOT_TEX4:
- case VARYING_SLOT_TEX5:
- case VARYING_SLOT_TEX6:
- case VARYING_SLOT_TEX7:
- if (st->needs_texcoord_semantic) {
- input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
- input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
- break;
- }
- /* fall through */
- case VARYING_SLOT_VAR0:
- default:
- assert(attr >= VARYING_SLOT_VAR0 ||
- (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
- input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- input_semantic_index[slot] =
- st_get_generic_varying_index(st, attr);
- break;
- }
- }
+ inputMapping[attr] = slot;
+ inputSlotToAttr[slot] = attr;
+
+ unsigned semantic_name, semantic_index;
+ tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
+ &semantic_name, &semantic_index);
+ input_semantic_name[slot] = semantic_name;
+ input_semantic_index[slot] = semantic_index;
}
/* Also add patch inputs. */
GLuint slot = num_outputs++;
outputMapping[attr] = slot;
- outputSlotToAttr[slot] = attr;
- switch (attr) {
- case VARYING_SLOT_POS:
- assert(slot == 0);
- output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_COL0:
- output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_COL1:
- output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- output_semantic_index[slot] = 1;
- break;
- case VARYING_SLOT_BFC0:
- output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_BFC1:
- output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
- output_semantic_index[slot] = 1;
- break;
- case VARYING_SLOT_FOGC:
- output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_PSIZ:
- output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_CLIP_VERTEX:
- output_semantic_name[slot] = TGSI_SEMANTIC_CLIPVERTEX;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_CLIP_DIST0:
- output_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_CLIP_DIST1:
- output_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
- output_semantic_index[slot] = 1;
- break;
- case VARYING_SLOT_CULL_DIST0:
- case VARYING_SLOT_CULL_DIST1:
- /* these should have been lowered by GLSL */
- assert(0);
- break;
- case VARYING_SLOT_LAYER:
- output_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_PRIMITIVE_ID:
- output_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_VIEWPORT:
- output_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_TESS_LEVEL_OUTER:
- output_semantic_name[slot] = TGSI_SEMANTIC_TESSOUTER;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_TESS_LEVEL_INNER:
- output_semantic_name[slot] = TGSI_SEMANTIC_TESSINNER;
- output_semantic_index[slot] = 0;
- break;
- case VARYING_SLOT_TEX0:
- case VARYING_SLOT_TEX1:
- case VARYING_SLOT_TEX2:
- case VARYING_SLOT_TEX3:
- case VARYING_SLOT_TEX4:
- case VARYING_SLOT_TEX5:
- case VARYING_SLOT_TEX6:
- case VARYING_SLOT_TEX7:
- if (st->needs_texcoord_semantic) {
- output_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
- output_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
- break;
- }
- /* fall through */
- case VARYING_SLOT_VAR0:
- default:
- assert(slot < ARRAY_SIZE(output_semantic_name));
- assert(attr >= VARYING_SLOT_VAR0 ||
- (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
- output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- output_semantic_index[slot] =
- st_get_generic_varying_index(st, attr);
- break;
- }
+ unsigned semantic_name, semantic_index;
+ tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
+ &semantic_name, &semantic_index);
+ output_semantic_name[slot] = semantic_name;
+ output_semantic_index[slot] = semantic_index;
}
}
GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
outputMapping[patch_attr] = slot;
- outputSlotToAttr[slot] = patch_attr;
output_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
output_semantic_index[slot] = attr;
}
/* outputs */
num_outputs,
outputMapping,
- outputSlotToAttr,
output_semantic_name,
output_semantic_index);
- out_state->tokens = ureg_get_tokens(ureg, NULL);
+ if (tgsi_processor == PIPE_SHADER_COMPUTE) {
+ struct st_compute_program *stcp = (struct st_compute_program *) prog;
+ out_state->tokens = ureg_get_tokens(ureg, &stcp->num_tgsi_tokens);
+ stcp->tgsi.prog = out_state->tokens;
+ } else {
+ struct st_common_program *stcp = (struct st_common_program *) prog;
+ out_state->tokens = ureg_get_tokens(ureg, &stcp->num_tgsi_tokens);
+ }
ureg_destroy(ureg);
- st_translate_stream_output_info(glsl_to_tgsi,
+ st_translate_stream_output_info(prog->sh.LinkedTransformFeedback,
outputMapping,
&out_state->stream_output);
+ st_store_ir_in_disk_cache(st, prog, false);
+
if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
_mesa_print_program(prog);
debug_printf("\n");
}
}
+/**
+ * Update stream-output info for GS/TCS/TES. Normally this is done in
+ * st_translate_program_common() but that is not called for glsl_to_nir
+ * case.
+ */
+static void
+st_translate_program_stream_output(struct gl_program *prog,
+ struct pipe_stream_output_info *stream_output)
+{
+ if (!prog->sh.LinkedTransformFeedback)
+ return;
+
+ ubyte outputMapping[VARYING_SLOT_TESS_MAX];
+ GLuint attr;
+ uint num_outputs = 0;
+
+ memset(outputMapping, 0, sizeof(outputMapping));
+
+ /*
+ * Determine number of outputs, the (default) output register
+ * mapping and the semantic information for each output.
+ */
+ for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
+ if (prog->info.outputs_written & BITFIELD64_BIT(attr)) {
+ GLuint slot = num_outputs++;
+
+ outputMapping[attr] = slot;
+ }
+ }
+
+ st_translate_stream_output_info(prog->sh.LinkedTransformFeedback,
+ outputMapping,
+ stream_output);
+}
/**
* Translate a geometry program to create a new variant.
*/
bool
st_translate_geometry_program(struct st_context *st,
- struct st_geometry_program *stgp)
+ struct st_common_program *stgp)
{
struct ureg_program *ureg;
+ /* We have already compiled to NIR so just return */
+ if (stgp->shader_program) {
+ /* No variants */
+ st_finalize_nir(st, &stgp->Base, stgp->shader_program,
+ stgp->tgsi.ir.nir);
+ st_translate_program_stream_output(&stgp->Base, &stgp->tgsi.stream_output);
+ st_store_ir_in_disk_cache(st, &stgp->Base, true);
+ return true;
+ }
+
ureg = ureg_create_with_screen(PIPE_SHADER_GEOMETRY, st->pipe->screen);
if (ureg == NULL)
return false;
struct st_basic_variant *
st_get_basic_variant(struct st_context *st,
unsigned pipe_shader,
- struct pipe_shader_state *tgsi,
- struct st_basic_variant **variants)
+ struct st_common_program *prog)
{
struct pipe_context *pipe = st->pipe;
struct st_basic_variant *v;
struct st_basic_variant_key key;
-
+ struct pipe_shader_state tgsi = {0};
memset(&key, 0, sizeof(key));
key.st = st->has_shareable_shaders ? NULL : st;
/* Search for existing variant */
- for (v = *variants; v; v = v->next) {
+ for (v = prog->variants; v; v = v->next) {
if (memcmp(&v->key, &key, sizeof(key)) == 0) {
break;
}
/* create new */
v = CALLOC_STRUCT(st_basic_variant);
if (v) {
+
+ if (prog->tgsi.type == PIPE_SHADER_IR_NIR) {
+ tgsi.type = PIPE_SHADER_IR_NIR;
+ tgsi.ir.nir = nir_shader_clone(NULL, prog->tgsi.ir.nir);
+ tgsi.stream_output = prog->tgsi.stream_output;
+ } else
+ tgsi = prog->tgsi;
/* fill in new variant */
switch (pipe_shader) {
case PIPE_SHADER_TESS_CTRL:
- v->driver_shader = pipe->create_tcs_state(pipe, tgsi);
+ v->driver_shader = pipe->create_tcs_state(pipe, &tgsi);
break;
case PIPE_SHADER_TESS_EVAL:
- v->driver_shader = pipe->create_tes_state(pipe, tgsi);
+ v->driver_shader = pipe->create_tes_state(pipe, &tgsi);
break;
case PIPE_SHADER_GEOMETRY:
- v->driver_shader = pipe->create_gs_state(pipe, tgsi);
+ v->driver_shader = pipe->create_gs_state(pipe, &tgsi);
break;
default:
assert(!"unhandled shader type");
v->key = key;
/* insert into list */
- v->next = *variants;
- *variants = v;
+ v->next = prog->variants;
+ prog->variants = v;
}
}
*/
bool
st_translate_tessctrl_program(struct st_context *st,
- struct st_tessctrl_program *sttcp)
+ struct st_common_program *sttcp)
{
struct ureg_program *ureg;
+ /* We have already compiled to NIR so just return */
+ if (sttcp->shader_program) {
+ /* No variants */
+ st_finalize_nir(st, &sttcp->Base, sttcp->shader_program,
+ sttcp->tgsi.ir.nir);
+ st_store_ir_in_disk_cache(st, &sttcp->Base, true);
+ return true;
+ }
+
ureg = ureg_create_with_screen(PIPE_SHADER_TESS_CTRL, st->pipe->screen);
if (ureg == NULL)
return false;
*/
bool
st_translate_tesseval_program(struct st_context *st,
- struct st_tesseval_program *sttep)
+ struct st_common_program *sttep)
{
struct ureg_program *ureg;
+ /* We have already compiled to NIR so just return */
+ if (sttep->shader_program) {
+ /* No variants */
+ st_finalize_nir(st, &sttep->Base, sttep->shader_program,
+ sttep->tgsi.ir.nir);
+ st_translate_program_stream_output(&sttep->Base, &sttep->tgsi.stream_output);
+ st_store_ir_in_disk_cache(st, &sttep->Base, true);
+ return true;
+ }
+
ureg = ureg_create_with_screen(PIPE_SHADER_TESS_EVAL, st->pipe->screen);
if (ureg == NULL)
return false;
struct ureg_program *ureg;
struct pipe_shader_state prog;
+ stcp->tgsi.req_local_mem = stcp->Base.info.cs.shared_size;
+
+ if (stcp->shader_program) {
+ /* no compute variants: */
+ st_finalize_nir(st, &stcp->Base, stcp->shader_program,
+ (struct nir_shader *) stcp->tgsi.prog);
+ st_store_ir_in_disk_cache(st, &stcp->Base, true);
+ return true;
+ }
+
ureg = ureg_create_with_screen(PIPE_SHADER_COMPUTE, st->pipe->screen);
if (ureg == NULL)
return false;
PIPE_SHADER_COMPUTE, &prog);
stcp->tgsi.ir_type = PIPE_SHADER_IR_TGSI;
- stcp->tgsi.prog = prog.tokens;
- stcp->tgsi.req_local_mem = stcp->Base.info.cs.shared_size;
stcp->tgsi.req_private_mem = 0;
stcp->tgsi.req_input_mem = 0;
struct st_basic_variant *v;
struct st_basic_variant_key key;
+ /* use memset, not an initializer to be sure all memory is zeroed */
memset(&key, 0, sizeof(key));
+
key.st = st->has_shareable_shaders ? NULL : st;
/* Search for existing variant */
v = CALLOC_STRUCT(st_basic_variant);
if (v) {
/* fill in new variant */
- v->driver_shader = pipe->create_compute_state(pipe, tgsi);
+ struct pipe_compute_state cs = *tgsi;
+ if (tgsi->ir_type == PIPE_SHADER_IR_NIR)
+ cs.prog = nir_shader_clone(NULL, tgsi->prog);
+ v->driver_shader = pipe->create_compute_state(pipe, &cs);
v->key = key;
/* insert into list */
case GL_TESS_EVALUATION_PROGRAM_NV:
case GL_COMPUTE_PROGRAM_NV:
{
- struct st_geometry_program *gp = (struct st_geometry_program*)target;
- struct st_tessctrl_program *tcp = (struct st_tessctrl_program*)target;
- struct st_tesseval_program *tep = (struct st_tesseval_program*)target;
+ struct st_common_program *p = st_common_program(target);
struct st_compute_program *cp = (struct st_compute_program*)target;
struct st_basic_variant **variants =
- target->Target == GL_GEOMETRY_PROGRAM_NV ? &gp->variants :
- target->Target == GL_TESS_CONTROL_PROGRAM_NV ? &tcp->variants :
- target->Target == GL_TESS_EVALUATION_PROGRAM_NV ? &tep->variants :
target->Target == GL_COMPUTE_PROGRAM_NV ? &cp->variants :
- NULL;
+ &p->variants;
struct st_basic_variant *v, **prevPtr = variants;
for (v = *variants; v; ) {
struct st_vp_variant_key key;
memset(&key, 0, sizeof(key));
+
key.st = st->has_shareable_shaders ? NULL : st;
st_get_vp_variant(st, p, &key);
break;
}
case GL_TESS_CONTROL_PROGRAM_NV: {
- struct st_tessctrl_program *p = (struct st_tessctrl_program *)prog;
- st_get_basic_variant(st, PIPE_SHADER_TESS_CTRL, &p->tgsi, &p->variants);
+ struct st_common_program *p = st_common_program(prog);
+ st_get_basic_variant(st, PIPE_SHADER_TESS_CTRL, p);
break;
}
case GL_TESS_EVALUATION_PROGRAM_NV: {
- struct st_tesseval_program *p = (struct st_tesseval_program *)prog;
- st_get_basic_variant(st, PIPE_SHADER_TESS_EVAL, &p->tgsi, &p->variants);
+ struct st_common_program *p = st_common_program(prog);
+ st_get_basic_variant(st, PIPE_SHADER_TESS_EVAL, p);
break;
}
case GL_GEOMETRY_PROGRAM_NV: {
- struct st_geometry_program *p = (struct st_geometry_program *)prog;
- st_get_basic_variant(st, PIPE_SHADER_GEOMETRY, &p->tgsi, &p->variants);
+ struct st_common_program *p = st_common_program(prog);
+ st_get_basic_variant(st, PIPE_SHADER_GEOMETRY, p);
break;
}
struct st_fp_variant_key key;
memset(&key, 0, sizeof(key));
+
key.st = st->has_shareable_shaders ? NULL : st;
st_get_fp_variant(st, p, &key);
break;
projects / mesa.git / blobdiff