#include "program/prog_statevars.h"
#include "program/prog_parameter.h"
#include "program/ir_to_mesa.h"
-#include "program/hash_table.h"
#include "main/mtypes.h"
#include "main/errors.h"
#include "main/shaderapi.h"
#include "st_context.h"
#include "st_program.h"
-#include "st_glsl_types.h"
#include "compiler/nir/nir.h"
-#include "compiler/nir/glsl_to_nir.h"
#include "compiler/glsl_types.h"
+#include "compiler/glsl/glsl_to_nir.h"
#include "compiler/glsl/ir.h"
+#include "compiler/glsl/string_to_uint_map.h"
+static int
+type_size(const struct glsl_type *type)
+{
+ return type->count_attribute_slots(false);
+}
+
/* Depending on PIPE_CAP_TGSI_TEXCOORD (st->needs_texcoord_semantic) we
* may need to fix up varying slots so the glsl->nir path is aligned
* with the anything->tgsi->nir path.
* on varying-slot w/ the VS outputs)
*/
static void
-st_nir_assign_vs_in_locations(struct gl_program *prog,
- struct exec_list *var_list, unsigned *size)
+st_nir_assign_vs_in_locations(struct gl_program *prog, nir_shader *nir)
{
unsigned attr, num_inputs = 0;
unsigned input_to_index[VERT_ATTRIB_MAX] = {0};
/* TODO de-duplicate w/ similar code in st_translate_vertex_program()? */
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
- if ((prog->InputsRead & BITFIELD64_BIT(attr)) != 0) {
+ if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
input_to_index[attr] = num_inputs;
num_inputs++;
- if ((prog->DoubleInputsRead & BITFIELD64_BIT(attr)) != 0) {
+ if ((prog->info.double_inputs_read & BITFIELD64_BIT(attr)) != 0) {
/* add placeholder for second part of a double attribute */
num_inputs++;
}
+ } else {
+ input_to_index[attr] = ~0;
}
}
- *size = 0;
- nir_foreach_variable(var, var_list) {
+ /* bit of a hack, mirroring st_translate_vertex_program */
+ input_to_index[VERT_ATTRIB_EDGEFLAG] = num_inputs;
+
+ nir->num_inputs = 0;
+ nir_foreach_variable_safe(var, &nir->inputs) {
attr = var->data.location;
assert(attr < ARRAY_SIZE(input_to_index));
- var->data.driver_location = input_to_index[attr];
- (*size)++;
+
+ if (input_to_index[attr] != ~0u) {
+ var->data.driver_location = input_to_index[attr];
+ nir->num_inputs++;
+ } else {
+ /* Move unused input variables to the globals list (with no
+ * initialization), to avoid confusing drivers looking through the
+ * inputs array and expecting to find inputs with a driver_location
+ * set.
+ */
+ exec_node_remove(&var->node);
+ var->data.mode = nir_var_global;
+ exec_list_push_tail(&nir->globals, &var->node);
+ }
}
}
+static void
+st_nir_assign_var_locations(struct exec_list *var_list, unsigned *size)
+{
+ unsigned location = 0;
+ unsigned assigned_locations[VARYING_SLOT_MAX];
+ uint64_t processed_locs = 0;
+
+ nir_foreach_variable(var, var_list) {
+ /* Because component packing allows varyings to share the same location
+ * we may have already have processed this location.
+ */
+ if (var->data.location >= VARYING_SLOT_VAR0 &&
+ processed_locs & ((uint64_t)1 << var->data.location)) {
+ var->data.driver_location = assigned_locations[var->data.location];
+ *size += type_size(var->type);
+ continue;
+ }
+
+ assigned_locations[var->data.location] = location;
+ var->data.driver_location = location;
+ location += type_size(var->type);
+
+ processed_locs |= ((uint64_t)1 << var->data.location);
+ }
+
+ *size += location;
+}
+
static int
st_nir_lookup_parameter_index(const struct gl_program_parameter_list *params,
const char *name)
{
int max = 0;
int shaderidx = 0;
+ int imageidx = 0;
nir_foreach_variable(uniform, uniform_list) {
int loc;
uniform->interface_type != NULL)
continue;
- if (uniform->type->is_sampler()) {
- unsigned val;
+ if (uniform->type->is_sampler() || uniform->type->is_image()) {
+ unsigned val = 0;
bool found = shader_program->UniformHash->get(val, uniform->name);
- loc = shaderidx++;
+ if (uniform->type->is_sampler())
+ loc = shaderidx++;
+ else
+ loc = imageidx++;
assert(found);
+ (void) found; /* silence unused var warning */
/* this ensure that nir_lower_samplers looks at the correct
* shader_program->UniformStorage[location]:
*/
uniform->data.driver_location = loc;
- max = MAX2(max, loc + st_glsl_type_size(uniform->type));
+ max = MAX2(max, loc + type_size(uniform->type));
}
*size = max;
}
-extern "C" {
+static void
+st_nir_opts(nir_shader *nir)
+{
+ bool progress;
+ do {
+ progress = false;
+
+ NIR_PASS(progress, nir, nir_copy_prop);
+ NIR_PASS(progress, nir, nir_opt_remove_phis);
+ NIR_PASS(progress, nir, nir_opt_dce);
+ if (nir_opt_trivial_continues(nir)) {
+ progress = true;
+ NIR_PASS(progress, nir, nir_copy_prop);
+ NIR_PASS(progress, nir, nir_opt_dce);
+ }
+ NIR_PASS(progress, nir, nir_opt_if);
+ NIR_PASS(progress, nir, nir_opt_dead_cf);
+ NIR_PASS(progress, nir, nir_opt_cse);
+ NIR_PASS(progress, nir, nir_opt_peephole_select, 8);
+
+ NIR_PASS(progress, nir, nir_opt_algebraic);
+ NIR_PASS(progress, nir, nir_opt_constant_folding);
+
+ NIR_PASS(progress, nir, nir_opt_undef);
+ NIR_PASS(progress, nir, nir_opt_conditional_discard);
+ if (nir->options->max_unroll_iterations) {
+ NIR_PASS(progress, nir, nir_opt_loop_unroll, (nir_variable_mode)0);
+ }
+ } while (progress);
+}
-/* First half of converting glsl_to_nir.. this leaves things in a pre-
+/* First third of converting glsl_to_nir.. this leaves things in a pre-
* nir_lower_io state, so that shader variants can more easily insert/
* replace variables, etc.
*/
-nir_shader *
+static nir_shader *
st_glsl_to_nir(struct st_context *st, struct gl_program *prog,
struct gl_shader_program *shader_program,
gl_shader_stage stage)
{
struct pipe_screen *pscreen = st->pipe->screen;
- unsigned ptarget = st_shader_stage_to_ptarget(stage);
+ enum pipe_shader_type ptarget = pipe_shader_type_from_mesa(stage);
const nir_shader_compiler_options *options;
- nir_shader *nir;
assert(pscreen->get_compiler_options); /* drivers using NIR must implement this */
if (prog->nir)
return prog->nir;
- nir = glsl_to_nir(shader_program, stage, options);
- prog->nir = nir;
+ nir_shader *nir = glsl_to_nir(shader_program, stage, options);
- NIR_PASS_V(nir, nir_lower_io_to_temporaries,
- nir_shader_get_entrypoint(nir),
- true, true);
- NIR_PASS_V(nir, nir_lower_global_vars_to_local);
- NIR_PASS_V(nir, nir_split_var_copies);
- NIR_PASS_V(nir, nir_lower_var_copies);
- NIR_PASS_V(nir, st_nir_lower_builtin);
+ st_nir_opts(nir);
+
+ return nir;
+}
- /* fragment shaders may need : */
- if (stage == MESA_SHADER_FRAGMENT) {
- static const gl_state_index wposTransformState[STATE_LENGTH] = {
- STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM
- };
- nir_lower_wpos_ytransform_options wpos_options = {0};
- struct pipe_screen *pscreen = st->pipe->screen;
-
- memcpy(wpos_options.state_tokens, wposTransformState,
- sizeof(wpos_options.state_tokens));
- wpos_options.fs_coord_origin_upper_left =
- pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT);
- wpos_options.fs_coord_origin_lower_left =
- pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
- wpos_options.fs_coord_pixel_center_integer =
- pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
- wpos_options.fs_coord_pixel_center_half_integer =
- pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER);
-
- if (nir_lower_wpos_ytransform(nir, &wpos_options)) {
- nir_validate_shader(nir);
- _mesa_add_state_reference(prog->Parameters, wposTransformState);
+/* Second third of converting glsl_to_nir. This creates uniforms, gathers
+ * info on varyings, etc after NIR link time opts have been applied.
+ */
+static void
+st_glsl_to_nir_post_opts(struct st_context *st, struct gl_program *prog,
+ struct gl_shader_program *shader_program)
+{
+ nir_shader *nir = prog->nir;
+
+ /* Make a pass over the IR to add state references for any built-in
+ * uniforms that are used. This has to be done now (during linking).
+ * Code generation doesn't happen until the first time this shader is
+ * used for rendering. Waiting until then to generate the parameters is
+ * too late. At that point, the values for the built-in uniforms won't
+ * get sent to the shader.
+ */
+ nir_foreach_variable(var, &nir->uniforms) {
+ if (strncmp(var->name, "gl_", 3) == 0) {
+ const nir_state_slot *const slots = var->state_slots;
+ assert(var->state_slots != NULL);
+
+ for (unsigned int i = 0; i < var->num_state_slots; i++) {
+ _mesa_add_state_reference(prog->Parameters,
+ (gl_state_index *)slots[i].tokens);
+ }
}
}
+ /* Avoid reallocation of the program parameter list, because the uniform
+ * storage is only associated with the original parameter list.
+ * This should be enough for Bitmap and DrawPixels constants.
+ */
+ _mesa_reserve_parameter_storage(prog->Parameters, 8);
+
+ /* This has to be done last. Any operation the can cause
+ * prog->ParameterValues to get reallocated (e.g., anything that adds a
+ * program constant) has to happen before creating this linkage.
+ */
+ _mesa_associate_uniform_storage(st->ctx, shader_program, prog, true);
+
+ st_set_prog_affected_state_flags(prog);
+
+ NIR_PASS_V(nir, st_nir_lower_builtin);
+ NIR_PASS_V(nir, nir_lower_atomics, shader_program);
+
if (st->ctx->_Shader->Flags & GLSL_DUMP) {
_mesa_log("\n");
_mesa_log("NIR IR for linked %s program %d:\n",
- _mesa_shader_stage_to_string(stage),
+ _mesa_shader_stage_to_string(prog->info.stage),
shader_program->Name);
nir_print_shader(nir, _mesa_get_log_file());
_mesa_log("\n\n");
}
-
- return nir;
}
/* TODO any better helper somewhere to sort a list? */
exec_list_move_nodes_to(&new_list, var_list);
}
-/* Second half of preparing nir from glsl, which happens after shader
- * variant lowering.
- */
-void
-st_finalize_nir(struct st_context *st, struct gl_program *prog, nir_shader *nir)
+static void
+set_st_program(struct gl_program *prog,
+ struct gl_shader_program *shader_program,
+ nir_shader *nir)
{
- NIR_PASS_V(nir, nir_split_var_copies);
- NIR_PASS_V(nir, nir_lower_var_copies);
- NIR_PASS_V(nir, nir_lower_io_types);
-
- if (nir->stage == MESA_SHADER_VERTEX) {
- /* Needs special handling so drvloc matches the vbo state: */
- st_nir_assign_vs_in_locations(prog, &nir->inputs, &nir->num_inputs);
- sort_varyings(&nir->outputs);
- nir_assign_var_locations(&nir->outputs,
- &nir->num_outputs,
- st_glsl_type_size);
- st_nir_fixup_varying_slots(st, &nir->outputs);
- } else if (nir->stage == MESA_SHADER_FRAGMENT) {
- sort_varyings(&nir->inputs);
- nir_assign_var_locations(&nir->inputs,
- &nir->num_inputs,
- st_glsl_type_size);
- st_nir_fixup_varying_slots(st, &nir->inputs);
- nir_assign_var_locations(&nir->outputs,
- &nir->num_outputs,
- st_glsl_type_size);
- } else {
- unreachable("invalid shader type for tgsi bypass\n");
- }
+ struct st_vertex_program *stvp;
+ struct st_common_program *stp;
+ struct st_fragment_program *stfp;
+ struct st_compute_program *stcp;
- struct gl_shader_program *shader_program;
- switch (nir->stage) {
+ switch (prog->info.stage) {
case MESA_SHADER_VERTEX:
- shader_program = ((struct st_vertex_program *)prog)->shader_program;
+ stvp = (struct st_vertex_program *)prog;
+ stvp->shader_program = shader_program;
+ stvp->tgsi.type = PIPE_SHADER_IR_NIR;
+ stvp->tgsi.ir.nir = nir;
+ break;
+ case MESA_SHADER_GEOMETRY:
+ case MESA_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_EVAL:
+ stp = (struct st_common_program *)prog;
+ stp->shader_program = shader_program;
+ stp->tgsi.type = PIPE_SHADER_IR_NIR;
+ stp->tgsi.ir.nir = nir;
break;
case MESA_SHADER_FRAGMENT:
- shader_program = ((struct st_fragment_program *)prog)->shader_program;
+ stfp = (struct st_fragment_program *)prog;
+ stfp->shader_program = shader_program;
+ stfp->tgsi.type = PIPE_SHADER_IR_NIR;
+ stfp->tgsi.ir.nir = nir;
+ break;
+ case MESA_SHADER_COMPUTE:
+ stcp = (struct st_compute_program *)prog;
+ stcp->shader_program = shader_program;
+ stcp->tgsi.ir_type = PIPE_SHADER_IR_NIR;
+ stcp->tgsi.prog = nir_shader_clone(NULL, nir);
break;
default:
- assert(!"should not be reached");
- return;
+ unreachable("unknown shader stage");
}
-
- st_nir_assign_uniform_locations(prog, shader_program,
- &nir->uniforms, &nir->num_uniforms);
-
- NIR_PASS_V(nir, nir_lower_system_values);
- NIR_PASS_V(nir, nir_lower_io, nir_var_all, st_glsl_type_size);
- NIR_PASS_V(nir, nir_lower_samplers, shader_program);
}
-struct gl_program *
+static void
st_nir_get_mesa_program(struct gl_context *ctx,
struct gl_shader_program *shader_program,
- struct gl_shader *shader)
+ struct gl_linked_shader *shader)
{
+ struct st_context *st = st_context(ctx);
struct gl_program *prog;
- GLenum target = _mesa_shader_stage_to_program(shader->Stage);
validate_ir_tree(shader->ir);
- prog = ctx->Driver.NewProgram(ctx, target, shader_program->Name);
- if (!prog)
- return NULL;
+ prog = shader->Program;
prog->Parameters = _mesa_new_parameter_list();
- _mesa_copy_linked_program_data(shader->Stage, shader_program, prog);
- _mesa_generate_parameters_list_for_uniforms(shader_program, shader,
+ _mesa_copy_linked_program_data(shader_program, shader);
+ _mesa_generate_parameters_list_for_uniforms(ctx, shader_program, shader,
prog->Parameters);
- /* Make a pass over the IR to add state references for any built-in
- * uniforms that are used. This has to be done now (during linking).
- * Code generation doesn't happen until the first time this shader is
- * used for rendering. Waiting until then to generate the parameters is
- * too late. At that point, the values for the built-in uniforms won't
- * get sent to the shader.
- */
- foreach_in_list(ir_instruction, node, shader->ir) {
- ir_variable *var = node->as_variable();
-
- if ((var == NULL) || (var->data.mode != ir_var_uniform) ||
- (strncmp(var->name, "gl_", 3) != 0))
- continue;
-
- const ir_state_slot *const slots = var->get_state_slots();
- assert(slots != NULL);
-
- for (unsigned int i = 0; i < var->get_num_state_slots(); i++) {
- _mesa_add_state_reference(prog->Parameters,
- (gl_state_index *) slots[i].tokens);
- }
- }
-
if (ctx->_Shader->Flags & GLSL_DUMP) {
_mesa_log("\n");
_mesa_log("GLSL IR for linked %s program %d:\n",
_mesa_log("\n\n");
}
- prog->Instructions = NULL;
- prog->NumInstructions = 0;
+ prog->ExternalSamplersUsed = gl_external_samplers(prog);
+ _mesa_update_shader_textures_used(shader_program, prog);
- do_set_program_inouts(shader->ir, prog, shader->Stage);
+ nir_shader *nir = st_glsl_to_nir(st, prog, shader_program, shader->Stage);
- prog->SamplersUsed = shader->active_samplers;
- prog->ShadowSamplers = shader->shadow_samplers;
- _mesa_update_shader_textures_used(shader_program, prog);
+ set_st_program(prog, shader_program, nir);
+ prog->nir = nir;
- _mesa_reference_program(ctx, &shader->Program, prog);
+ NIR_PASS_V(nir, nir_lower_io_to_temporaries,
+ nir_shader_get_entrypoint(nir),
+ true, true);
+ NIR_PASS_V(nir, nir_lower_global_vars_to_local);
+ NIR_PASS_V(nir, nir_split_var_copies);
+ NIR_PASS_V(nir, nir_lower_var_copies);
+}
- /* Avoid reallocation of the program parameter list, because the uniform
- * storage is only associated with the original parameter list.
- * This should be enough for Bitmap and DrawPixels constants.
- */
- _mesa_reserve_parameter_storage(prog->Parameters, 8);
+extern "C" {
- /* This has to be done last. Any operation the can cause
- * prog->ParameterValues to get reallocated (e.g., anything that adds a
- * program constant) has to happen before creating this linkage.
- */
- _mesa_associate_uniform_storage(ctx, shader_program, prog->Parameters);
+bool
+st_link_nir(struct gl_context *ctx,
+ struct gl_shader_program *shader_program)
+{
+ struct st_context *st = st_context(ctx);
- struct st_vertex_program *stvp;
- struct st_fragment_program *stfp;
+ for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
+ struct gl_linked_shader *shader = shader_program->_LinkedShaders[i];
+ if (shader == NULL)
+ continue;
- switch (shader->Type) {
- case GL_VERTEX_SHADER:
- stvp = (struct st_vertex_program *)prog;
- stvp->shader_program = shader_program;
- break;
- case GL_FRAGMENT_SHADER:
- stfp = (struct st_fragment_program *)prog;
- stfp->shader_program = shader_program;
- break;
- default:
- assert(!"should not be reached");
- return NULL;
+ st_nir_get_mesa_program(ctx, shader_program, shader);
+ }
+
+ for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
+ struct gl_linked_shader *shader = shader_program->_LinkedShaders[i];
+ if (shader == NULL)
+ continue;
+
+ nir_shader *nir = shader->Program->nir;
+
+ /* fragment shaders may need : */
+ if (nir->info.stage == MESA_SHADER_FRAGMENT) {
+ static const gl_state_index wposTransformState[STATE_LENGTH] = {
+ STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM
+ };
+ nir_lower_wpos_ytransform_options wpos_options = { { 0 } };
+ struct pipe_screen *pscreen = st->pipe->screen;
+
+ memcpy(wpos_options.state_tokens, wposTransformState,
+ sizeof(wpos_options.state_tokens));
+ wpos_options.fs_coord_origin_upper_left =
+ pscreen->get_param(pscreen,
+ PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT);
+ wpos_options.fs_coord_origin_lower_left =
+ pscreen->get_param(pscreen,
+ PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
+ wpos_options.fs_coord_pixel_center_integer =
+ pscreen->get_param(pscreen,
+ PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
+ wpos_options.fs_coord_pixel_center_half_integer =
+ pscreen->get_param(pscreen,
+ PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER);
+
+ if (nir_lower_wpos_ytransform(nir, &wpos_options)) {
+ nir_validate_shader(nir);
+ _mesa_add_state_reference(shader->Program->Parameters,
+ wposTransformState);
+ }
+ }
+
+ NIR_PASS_V(nir, nir_lower_system_values);
+
+ nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
+ shader->Program->info = nir->info;
+
+ st_glsl_to_nir_post_opts(st, shader->Program, shader_program);
+
+ assert(shader->Program);
+ if (!ctx->Driver.ProgramStringNotify(ctx,
+ _mesa_shader_stage_to_program(i),
+ shader->Program)) {
+ _mesa_reference_program(ctx, &shader->Program, NULL);
+ return false;
+ }
+ }
+
+ return true;
+}
+
+/* Last third of preparing nir from glsl, which happens after shader
+ * variant lowering.
+ */
+void
+st_finalize_nir(struct st_context *st, struct gl_program *prog,
+ struct gl_shader_program *shader_program, nir_shader *nir)
+{
+ struct pipe_screen *screen = st->pipe->screen;
+
+ NIR_PASS_V(nir, nir_split_var_copies);
+ NIR_PASS_V(nir, nir_lower_var_copies);
+ NIR_PASS_V(nir, nir_lower_io_types);
+
+ if (nir->info.stage == MESA_SHADER_VERTEX) {
+ /* Needs special handling so drvloc matches the vbo state: */
+ st_nir_assign_vs_in_locations(prog, nir);
+ /* Re-lower global vars, to deal with any dead VS inputs. */
+ NIR_PASS_V(nir, nir_lower_global_vars_to_local);
+
+ sort_varyings(&nir->outputs);
+ st_nir_assign_var_locations(&nir->outputs,
+ &nir->num_outputs);
+ st_nir_fixup_varying_slots(st, &nir->outputs);
+ } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
+ sort_varyings(&nir->inputs);
+ st_nir_assign_var_locations(&nir->inputs,
+ &nir->num_inputs);
+ st_nir_fixup_varying_slots(st, &nir->inputs);
+ st_nir_assign_var_locations(&nir->outputs,
+ &nir->num_outputs);
+ } else if (nir->info.stage == MESA_SHADER_COMPUTE) {
+ /* TODO? */
+ } else {
+ unreachable("invalid shader type for tgsi bypass\n");
}
- return prog;
+ NIR_PASS_V(nir, nir_lower_atomics_to_ssbo,
+ st->ctx->Const.Program[nir->info.stage].MaxAtomicBuffers);
+
+ st_nir_assign_uniform_locations(prog, shader_program,
+ &nir->uniforms, &nir->num_uniforms);
+
+ if (screen->get_param(screen, PIPE_CAP_NIR_SAMPLERS_AS_DEREF))
+ NIR_PASS_V(nir, nir_lower_samplers_as_deref, shader_program);
+ else
+ NIR_PASS_V(nir, nir_lower_samplers, shader_program);
}
} /* extern "C" */