int type; /**< GL_DOUBLE, GL_FLOAT, GL_INT, GL_BOOL, or GL_UNSIGNED_INT */
};
-static st_src_reg undef_src = st_src_reg(PROGRAM_UNDEFINED, 0, GLSL_TYPE_ERROR);
-static st_dst_reg undef_dst = st_dst_reg(PROGRAM_UNDEFINED, SWIZZLE_NOOP, GLSL_TYPE_ERROR);
+static const st_src_reg undef_src = st_src_reg(PROGRAM_UNDEFINED, 0, GLSL_TYPE_ERROR);
+static const st_dst_reg undef_dst = st_dst_reg(PROGRAM_UNDEFINED, SWIZZLE_NOOP, GLSL_TYPE_ERROR);
struct inout_decl {
unsigned mesa_index;
void visit_membar_intrinsic(ir_call *);
void visit_shared_intrinsic(ir_call *);
void visit_image_intrinsic(ir_call *);
+ void visit_generic_intrinsic(ir_call *, unsigned op);
st_src_reg result;
void rename_temp_registers(int num_renames, struct rename_reg_pair *renames);
void get_first_temp_read(int *first_reads);
+ void get_first_temp_write(int *first_writes);
void get_last_temp_read_first_temp_write(int *last_reads, int *first_writes);
void get_last_temp_write(int *last_writes);
const_offset % 16 / 4,
const_offset % 16 / 4);
- if (ir->type->base_type == GLSL_TYPE_BOOL) {
+ if (ir->type->is_boolean()) {
emit_asm(ir, TGSI_OPCODE_USNE, result_dst, cbuf, st_src_reg_for_int(0));
} else {
emit_asm(ir, TGSI_OPCODE_MOV, result_dst, cbuf);
break;
case ir_binop_ldexp:
- if (ir->operands[0]->type->base_type == GLSL_TYPE_DOUBLE) {
+ if (ir->operands[0]->type->is_double()) {
emit_asm(ir, TGSI_OPCODE_DLDEXP, result_dst, op[0], op[1]);
} else {
assert(!"Invalid ldexp for non-double opcode in glsl_to_tgsi_visitor::visit()");
break;
}
- case ir_unop_vote_any:
- emit_asm(ir, TGSI_OPCODE_VOTE_ANY, result_dst, op[0]);
- break;
- case ir_unop_vote_all:
- emit_asm(ir, TGSI_OPCODE_VOTE_ALL, result_dst, op[0]);
- break;
- case ir_unop_vote_eq:
- emit_asm(ir, TGSI_OPCODE_VOTE_EQ, result_dst, op[0]);
- break;
case ir_unop_u2i64:
case ir_unop_u2u64:
case ir_unop_b2i64: {
case ir_unop_unpack_snorm_4x8:
case ir_unop_unpack_unorm_4x8:
+ case ir_unop_unpack_sampler_2x32:
+ case ir_unop_pack_sampler_2x32:
+ case ir_unop_unpack_image_2x32:
+ case ir_unop_pack_image_2x32:
+
case ir_quadop_vector:
case ir_binop_vector_extract:
case ir_triop_vector_insert:
st_dst_reg *l, st_src_reg *r,
st_src_reg *cond, bool cond_swap)
{
- if (type->base_type == GLSL_TYPE_STRUCT) {
+ if (type->is_record()) {
for (unsigned int i = 0; i < type->length; i++) {
emit_block_mov(ir, type->fields.structure[i].type, l, r,
cond, cond_swap);
* aggregate constant and move each constant value into it. If we
* get lucky, copy propagation will eliminate the extra moves.
*/
- if (ir->type->base_type == GLSL_TYPE_STRUCT) {
+ if (ir->type->is_record()) {
st_src_reg temp_base = get_temp(ir->type);
st_dst_reg temp = st_dst_reg(temp_base);
inst->image_format = st_mesa_format_to_pipe_format(st_context(ctx),
_mesa_get_shader_image_format(imgvar->data.image_format));
- if (imgvar->data.image_coherent)
+ if (imgvar->data.memory_coherent)
inst->buffer_access |= TGSI_MEMORY_COHERENT;
- if (imgvar->data.image_restrict)
+ if (imgvar->data.memory_restrict)
inst->buffer_access |= TGSI_MEMORY_RESTRICT;
- if (imgvar->data.image_volatile)
+ if (imgvar->data.memory_volatile)
inst->buffer_access |= TGSI_MEMORY_VOLATILE;
}
+void
+glsl_to_tgsi_visitor::visit_generic_intrinsic(ir_call *ir, unsigned op)
+{
+ ir->return_deref->accept(this);
+ st_dst_reg dst = st_dst_reg(this->result);
+
+ st_src_reg src[4] = { undef_src, undef_src, undef_src, undef_src };
+ unsigned num_src = 0;
+ foreach_in_list(ir_rvalue, param, &ir->actual_parameters) {
+ assert(num_src < ARRAY_SIZE(src));
+
+ this->result.file = PROGRAM_UNDEFINED;
+ param->accept(this);
+ assert(this->result.file != PROGRAM_UNDEFINED);
+
+ src[num_src] = this->result;
+ num_src++;
+ }
+
+ emit_asm(ir, op, dst, src[0], src[1], src[2], src[3]);
+}
+
void
glsl_to_tgsi_visitor::visit(ir_call *ir)
{
visit_image_intrinsic(ir);
return;
- case ir_intrinsic_shader_clock: {
- ir->return_deref->accept(this);
-
- st_dst_reg dst = st_dst_reg(this->result);
- dst.writemask = TGSI_WRITEMASK_XY;
+ case ir_intrinsic_shader_clock:
+ visit_generic_intrinsic(ir, TGSI_OPCODE_CLOCK);
+ return;
- emit_asm(ir, TGSI_OPCODE_CLOCK, dst);
+ case ir_intrinsic_vote_all:
+ visit_generic_intrinsic(ir, TGSI_OPCODE_VOTE_ALL);
+ return;
+ case ir_intrinsic_vote_any:
+ visit_generic_intrinsic(ir, TGSI_OPCODE_VOTE_ANY);
+ return;
+ case ir_intrinsic_vote_eq:
+ visit_generic_intrinsic(ir, TGSI_OPCODE_VOTE_EQ);
+ return;
+ case ir_intrinsic_ballot:
+ visit_generic_intrinsic(ir, TGSI_OPCODE_BALLOT);
+ return;
+ case ir_intrinsic_read_first_invocation:
+ visit_generic_intrinsic(ir, TGSI_OPCODE_READ_FIRST);
+ return;
+ case ir_intrinsic_read_invocation:
+ visit_generic_intrinsic(ir, TGSI_OPCODE_READ_INVOC);
return;
- }
case ir_intrinsic_invalid:
case ir_intrinsic_generic_load:
component = this->result;
if (ir->offset) {
ir->offset->accept(this);
- if (ir->offset->type->base_type == GLSL_TYPE_ARRAY) {
+ if (ir->offset->type->is_array()) {
const glsl_type *elt_type = ir->offset->type->fields.array;
for (i = 0; i < ir->offset->type->length; i++) {
offset[i] = this->result;
/* Give up if we encounter relative addressing or flow control. */
if (inst->dst[0].reladdr || inst->dst[0].reladdr2 ||
inst->dst[1].reladdr || inst->dst[1].reladdr2 ||
- tgsi_get_opcode_info(inst->op)->is_branch ||
+ inst->info->is_branch ||
inst->op == TGSI_OPCODE_CONT ||
inst->op == TGSI_OPCODE_END ||
inst->op == TGSI_OPCODE_RET) {
}
}
+void
+glsl_to_tgsi_visitor::get_first_temp_write(int *first_writes)
+{
+ int depth = 0; /* loop depth */
+ int loop_start = -1; /* index of the first active BGNLOOP (if any) */
+ unsigned i = 0, j;
+
+ foreach_in_list(glsl_to_tgsi_instruction, inst, &this->instructions) {
+ for (j = 0; j < num_inst_dst_regs(inst); j++) {
+ if (inst->dst[j].file == PROGRAM_TEMPORARY) {
+ if (first_writes[inst->dst[j].index] == -1)
+ first_writes[inst->dst[j].index] = (depth == 0) ? i : loop_start;
+ }
+ }
+
+ if (inst->op == TGSI_OPCODE_BGNLOOP) {
+ if(depth++ == 0)
+ loop_start = i;
+ } else if (inst->op == TGSI_OPCODE_ENDLOOP) {
+ if (--depth == 0)
+ loop_start = -1;
+ }
+ assert(depth >= 0);
+ i++;
+ }
+}
+
void
glsl_to_tgsi_visitor::get_first_temp_read(int *first_reads)
{
void
glsl_to_tgsi_visitor::merge_registers(void)
{
- int *last_reads = rzalloc_array(mem_ctx, int, this->next_temp);
- int *first_writes = rzalloc_array(mem_ctx, int, this->next_temp);
+ int *last_reads = ralloc_array(mem_ctx, int, this->next_temp);
+ int *first_writes = ralloc_array(mem_ctx, int, this->next_temp);
struct rename_reg_pair *renames = rzalloc_array(mem_ctx, struct rename_reg_pair, this->next_temp);
int i, j;
int num_renames = 0;
{
int i = 0;
int new_index = 0;
- int *first_reads = rzalloc_array(mem_ctx, int, this->next_temp);
+ int *first_writes = ralloc_array(mem_ctx, int, this->next_temp);
struct rename_reg_pair *renames = rzalloc_array(mem_ctx, struct rename_reg_pair, this->next_temp);
int num_renames = 0;
+
for (i = 0; i < this->next_temp; i++) {
- first_reads[i] = -1;
+ first_writes[i] = -1;
}
- get_first_temp_read(first_reads);
+ get_first_temp_write(first_writes);
for (i = 0; i < this->next_temp; i++) {
- if (first_reads[i] < 0) continue;
+ if (first_writes[i] < 0) continue;
if (i != new_index) {
renames[num_renames].old_reg = i;
renames[num_renames].new_reg = new_index;
rename_temp_registers(num_renames, renames);
this->next_temp = new_index;
ralloc_free(renames);
- ralloc_free(first_reads);
+ ralloc_free(first_writes);
}
/* ------------------------- TGSI conversion stuff -------------------------- */
struct inout_decl *output_decls;
unsigned num_output_decls;
- const GLuint *inputMapping;
- const GLuint *outputMapping;
+ const ubyte *inputMapping;
+ const ubyte *outputMapping;
unsigned procType; /**< PIPE_SHADER_VERTEX/FRAGMENT */
};
case SYSTEM_VALUE_LOCAL_GROUP_SIZE:
return TGSI_SEMANTIC_BLOCK_SIZE;
+ /* ARB_shader_ballot */
+ case SYSTEM_VALUE_SUBGROUP_SIZE:
+ return TGSI_SEMANTIC_SUBGROUP_SIZE;
+ case SYSTEM_VALUE_SUBGROUP_INVOCATION:
+ return TGSI_SEMANTIC_SUBGROUP_INVOCATION;
+ case SYSTEM_VALUE_SUBGROUP_EQ_MASK:
+ return TGSI_SEMANTIC_SUBGROUP_EQ_MASK;
+ case SYSTEM_VALUE_SUBGROUP_GE_MASK:
+ return TGSI_SEMANTIC_SUBGROUP_GE_MASK;
+ case SYSTEM_VALUE_SUBGROUP_GT_MASK:
+ return TGSI_SEMANTIC_SUBGROUP_GT_MASK;
+ case SYSTEM_VALUE_SUBGROUP_LE_MASK:
+ return TGSI_SEMANTIC_SUBGROUP_LE_MASK;
+ case SYSTEM_VALUE_SUBGROUP_LT_MASK:
+ return TGSI_SEMANTIC_SUBGROUP_LT_MASK;
+
/* Unhandled */
case SYSTEM_VALUE_LOCAL_INVOCATION_INDEX:
case SYSTEM_VALUE_GLOBAL_INVOCATION_ID:
return mapping[a.mesa_index] < mapping[b.mesa_index];
}
- const GLuint *mapping;
+ const ubyte *mapping;
};
/* Sort the given array of decls by the corresponding slot (TGSI file index).
static void
sort_inout_decls_by_slot(struct inout_decl *decls,
unsigned count,
- const GLuint mapping[])
+ const ubyte mapping[])
{
sort_inout_decls sorter;
sorter.mapping = mapping;
glsl_to_tgsi_visitor *program,
const struct gl_program *proginfo,
GLuint numInputs,
- const GLuint inputMapping[],
- const GLuint inputSlotToAttr[],
+ const ubyte inputMapping[],
+ const ubyte inputSlotToAttr[],
const ubyte inputSemanticName[],
const ubyte inputSemanticIndex[],
- const GLuint interpMode[],
+ const ubyte interpMode[],
GLuint numOutputs,
- const GLuint outputMapping[],
- const GLuint outputSlotToAttr[],
+ const ubyte outputMapping[],
const ubyte outputSemanticName[],
const ubyte outputSemanticIndex[])
{
&ctx->Const.ShaderCompilerOptions[shader->Stage];
struct pipe_screen *pscreen = ctx->st->pipe->screen;
enum pipe_shader_type ptarget = st_shader_stage_to_ptarget(shader->Stage);
+ unsigned skip_merge_registers;
validate_ir_tree(shader->ir);
PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED);
v->has_tex_txf_lz = pscreen->get_param(pscreen,
PIPE_CAP_TGSI_TEX_TXF_LZ);
+ skip_merge_registers =
+ pscreen->get_shader_param(pscreen, ptarget,
+ PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS);
_mesa_generate_parameters_list_for_uniforms(shader_program, shader,
prog->Parameters);
* optimization passes. */
{
int i;
- int *first_writes = rzalloc_array(v->mem_ctx, int, v->next_temp);
- int *first_reads = rzalloc_array(v->mem_ctx, int, v->next_temp);
- int *last_writes = rzalloc_array(v->mem_ctx, int, v->next_temp);
- int *last_reads = rzalloc_array(v->mem_ctx, int, v->next_temp);
+ int *first_writes = ralloc_array(v->mem_ctx, int, v->next_temp);
+ int *first_reads = ralloc_array(v->mem_ctx, int, v->next_temp);
+ int *last_writes = ralloc_array(v->mem_ctx, int, v->next_temp);
+ int *last_reads = ralloc_array(v->mem_ctx, int, v->next_temp);
for (i = 0; i < v->next_temp; i++) {
first_writes[i] = -1;
while (v->eliminate_dead_code());
v->merge_two_dsts();
- v->merge_registers();
+ if (!skip_merge_registers)
+ v->merge_registers();
v->renumber_registers();
/* Write the END instruction. */
struct st_vertex_program *stvp;
struct st_fragment_program *stfp;
- struct st_geometry_program *stgp;
- struct st_tessctrl_program *sttcp;
- struct st_tesseval_program *sttep;
+ struct st_common_program *stgp;
+ struct st_common_program *sttcp;
+ struct st_common_program *sttep;
struct st_compute_program *stcp;
switch (shader->Stage) {
stfp->glsl_to_tgsi = v;
break;
case MESA_SHADER_GEOMETRY:
- stgp = (struct st_geometry_program *)prog;
+ stgp = (struct st_common_program *)prog;
stgp->glsl_to_tgsi = v;
break;
case MESA_SHADER_TESS_CTRL:
- sttcp = (struct st_tessctrl_program *)prog;
+ sttcp = (struct st_common_program *)prog;
sttcp->glsl_to_tgsi = v;
break;
case MESA_SHADER_TESS_EVAL:
- sttep = (struct st_tesseval_program *)prog;
+ sttep = (struct st_common_program *)prog;
sttep->glsl_to_tgsi = v;
break;
case MESA_SHADER_COMPUTE:
struct gl_program *linked_prog = NULL;
if (preferred_ir == PIPE_SHADER_IR_NIR) {
- /* TODO only for GLSL VS/FS for now: */
+ /* TODO only for GLSL VS/FS/CS for now: */
switch (shader->Stage) {
case MESA_SHADER_VERTEX:
case MESA_SHADER_FRAGMENT:
+ case MESA_SHADER_COMPUTE:
linked_prog = st_nir_get_mesa_program(ctx, prog, shader);
default:
break;
void
st_translate_stream_output_info(glsl_to_tgsi_visitor *glsl_to_tgsi,
- const GLuint outputMapping[],
+ const ubyte outputMapping[],
struct pipe_stream_output_info *so)
{
if (!glsl_to_tgsi->shader_program->last_vert_prog)
void
st_translate_stream_output_info2(struct gl_transform_feedback_info *info,
- const GLuint outputMapping[],
+ const ubyte outputMapping[],
struct pipe_stream_output_info *so)
{
unsigned i;
projects / mesa.git / blobdiff