#include "compiler/glsl/glsl_parser_extras.h"
#include "compiler/glsl/ir_optimization.h"
+#include "compiler/glsl/program.h"
#include "main/errors.h"
#include "main/shaderobj.h"
#include "main/uniforms.h"
#include "main/shaderapi.h"
+#include "main/shaderimage.h"
#include "program/prog_instruction.h"
-#include "program/sampler.h"
#include "pipe/p_context.h"
#include "pipe/p_screen.h"
#include "util/u_memory.h"
#include "st_program.h"
#include "st_mesa_to_tgsi.h"
+#include "st_format.h"
#define PROGRAM_ANY_CONST ((1 << PROGRAM_STATE_VAR) | \
this->index = index;
this->index2D = 0;
this->writemask = writemask;
- this->cond_mask = COND_TR;
this->reladdr = NULL;
this->reladdr2 = NULL;
this->has_index2 = false;
this->index = 0;
this->index2D = 0;
this->writemask = writemask;
- this->cond_mask = COND_TR;
this->reladdr = NULL;
this->reladdr2 = NULL;
this->has_index2 = false;
this->index = 0;
this->index2D = 0;
this->writemask = 0;
- this->cond_mask = COND_TR;
this->reladdr = NULL;
this->reladdr2 = NULL;
this->has_index2 = false;
int index; /**< temporary index, VERT_ATTRIB_*, VARYING_SLOT_*, etc. */
int index2D;
int writemask; /**< Bitfield of WRITEMASK_[XYZW] */
- GLuint cond_mask:4;
int type; /** GLSL_TYPE_* from GLSL IR (enum glsl_base_type) */
/** Register index should be offset by the integer in this reg. */
st_src_reg *reladdr;
this->file = reg.file;
this->index = reg.index;
this->writemask = WRITEMASK_XYZW;
- this->cond_mask = COND_TR;
this->reladdr = reg.reladdr;
this->index2D = reg.index2D;
this->reladdr2 = reg.reladdr2;
GLboolean cond_update;
bool saturate;
st_src_reg sampler; /**< sampler register */
+ int sampler_base;
int sampler_array_size; /**< 1-based size of sampler array, 1 if not array */
int tex_target; /**< One of TEXTURE_*_INDEX */
glsl_base_type tex_type;
GLboolean tex_shadow;
+ unsigned image_format;
st_src_reg tex_offsets[MAX_GLSL_TEXTURE_OFFSET];
unsigned tex_offset_num_offset;
glsl_base_type sampler_types[PIPE_MAX_SAMPLERS];
int sampler_targets[PIPE_MAX_SAMPLERS]; /**< One of TGSI_TEXTURE_* */
int buffers_used;
+ int images_used;
+ int image_targets[PIPE_MAX_SHADER_IMAGES];
+ unsigned image_formats[PIPE_MAX_SHADER_IMAGES];
bool indirect_addr_consts;
int wpos_transform_const;
bool native_integers;
bool have_sqrt;
bool have_fma;
+ bool use_shared_memory;
variable_storage *find_variable_storage(ir_variable *var);
void visit_atomic_counter_intrinsic(ir_call *);
void visit_ssbo_intrinsic(ir_call *);
void visit_membar_intrinsic(ir_call *);
+ void visit_shared_intrinsic(ir_call *);
+ void visit_image_intrinsic(ir_call *);
st_src_reg result;
void emit_arl(ir_instruction *ir, st_dst_reg dst, st_src_reg src0);
+ void get_deref_offsets(ir_dereference *ir,
+ unsigned *array_size,
+ unsigned *base,
+ unsigned *index,
+ st_src_reg *reladdr);
+ void calc_deref_offsets(ir_dereference *head,
+ ir_dereference *tail,
+ unsigned *array_elements,
+ unsigned *base,
+ unsigned *index,
+ st_src_reg *indirect,
+ unsigned *location);
+
bool try_emit_mad(ir_expression *ir,
int mul_operand);
bool try_emit_mad_for_and_not(ir_expression *ir,
case GLSL_TYPE_INTERFACE:
case GLSL_TYPE_VOID:
case GLSL_TYPE_ERROR:
+ case GLSL_TYPE_FUNCTION:
assert(!"Invalid type in type_size");
break;
}
case ir_unop_u2i:
/* Converting between signed and unsigned integers is a no-op. */
result_src = op[0];
+ result_src.type = result_dst.type;
break;
case ir_unop_b2i:
if (native_integers) {
switch (var->data.mode) {
case ir_var_uniform:
entry = new(mem_ctx) variable_storage(var, PROGRAM_UNIFORM,
- var->data.location);
+ var->data.param_index);
this->variables.push_tail(entry);
break;
case ir_var_shader_in:
glsl_to_tgsi_visitor::visit_atomic_counter_intrinsic(ir_call *ir)
{
const char *callee = ir->callee->function_name();
- ir_dereference *deref = static_cast<ir_dereference *>(
- ir->actual_parameters.get_head());
+ exec_node *param = ir->actual_parameters.get_head();
+ ir_dereference *deref = static_cast<ir_dereference *>(param);
ir_variable *location = deref->variable_referenced();
st_src_reg buffer(
/* Calculate the surface offset */
st_src_reg offset;
- ir_dereference_array *deref_array = deref->as_dereference_array();
-
- if (deref_array) {
- offset = get_temp(glsl_type::uint_type);
+ unsigned array_size = 0, base = 0, index = 0;
- deref_array->array_index->accept(this);
+ get_deref_offsets(deref, &array_size, &base, &index, &offset);
+ if (offset.file != PROGRAM_UNDEFINED) {
emit_asm(ir, TGSI_OPCODE_MUL, st_dst_reg(offset),
- this->result, st_src_reg_for_int(ATOMIC_COUNTER_SIZE));
+ offset, st_src_reg_for_int(ATOMIC_COUNTER_SIZE));
emit_asm(ir, TGSI_OPCODE_ADD, st_dst_reg(offset),
- offset, st_src_reg_for_int(location->data.offset));
+ offset, st_src_reg_for_int(location->data.offset + index * ATOMIC_COUNTER_SIZE));
} else {
- offset = st_src_reg_for_int(location->data.offset);
+ offset = st_src_reg_for_int(location->data.offset + index * ATOMIC_COUNTER_SIZE);
}
ir->return_deref->accept(this);
if (!strcmp("__intrinsic_atomic_read", callee)) {
inst = emit_asm(ir, TGSI_OPCODE_LOAD, dst, offset);
- inst->buffer = buffer;
} else if (!strcmp("__intrinsic_atomic_increment", callee)) {
inst = emit_asm(ir, TGSI_OPCODE_ATOMUADD, dst, offset,
st_src_reg_for_int(1));
- inst->buffer = buffer;
} else if (!strcmp("__intrinsic_atomic_predecrement", callee)) {
inst = emit_asm(ir, TGSI_OPCODE_ATOMUADD, dst, offset,
st_src_reg_for_int(-1));
- inst->buffer = buffer;
emit_asm(ir, TGSI_OPCODE_ADD, dst, this->result, st_src_reg_for_int(-1));
+ } else {
+ param = param->get_next();
+ ir_rvalue *val = ((ir_instruction *)param)->as_rvalue();
+ val->accept(this);
+
+ st_src_reg data = this->result, data2 = undef_src;
+ unsigned opcode;
+ if (!strcmp("__intrinsic_atomic_add", callee))
+ opcode = TGSI_OPCODE_ATOMUADD;
+ else if (!strcmp("__intrinsic_atomic_min", callee))
+ opcode = TGSI_OPCODE_ATOMIMIN;
+ else if (!strcmp("__intrinsic_atomic_max", callee))
+ opcode = TGSI_OPCODE_ATOMIMAX;
+ else if (!strcmp("__intrinsic_atomic_and", callee))
+ opcode = TGSI_OPCODE_ATOMAND;
+ else if (!strcmp("__intrinsic_atomic_or", callee))
+ opcode = TGSI_OPCODE_ATOMOR;
+ else if (!strcmp("__intrinsic_atomic_xor", callee))
+ opcode = TGSI_OPCODE_ATOMXOR;
+ else if (!strcmp("__intrinsic_atomic_exchange", callee))
+ opcode = TGSI_OPCODE_ATOMXCHG;
+ else if (!strcmp("__intrinsic_atomic_comp_swap", callee)) {
+ opcode = TGSI_OPCODE_ATOMCAS;
+ param = param->get_next();
+ val = ((ir_instruction *)param)->as_rvalue();
+ val->accept(this);
+ data2 = this->result;
+ } else if (!strcmp("__intrinsic_atomic_sub", callee)) {
+ opcode = TGSI_OPCODE_ATOMUADD;
+ st_src_reg res = get_temp(glsl_type::uvec4_type);
+ st_dst_reg dstres = st_dst_reg(res);
+ dstres.writemask = dst.writemask;
+ emit_asm(ir, TGSI_OPCODE_INEG, dstres, data);
+ data = res;
+ } else {
+ assert(!"Unexpected intrinsic");
+ return;
+ }
+
+ inst = emit_asm(ir, opcode, dst, offset, data, data2);
}
+
+ inst->buffer = buffer;
}
void
assert(!"Unexpected memory barrier intrinsic");
}
+void
+glsl_to_tgsi_visitor::visit_shared_intrinsic(ir_call *ir)
+{
+ const char *callee = ir->callee->function_name();
+ exec_node *param = ir->actual_parameters.get_head();
+
+ ir_rvalue *offset = ((ir_instruction *)param)->as_rvalue();
+
+ st_src_reg buffer(PROGRAM_MEMORY, 0, GLSL_TYPE_UINT);
+
+ /* Calculate the surface offset */
+ offset->accept(this);
+ st_src_reg off = this->result;
+
+ st_dst_reg dst = undef_dst;
+ if (ir->return_deref) {
+ ir->return_deref->accept(this);
+ dst = st_dst_reg(this->result);
+ dst.writemask = (1 << ir->return_deref->type->vector_elements) - 1;
+ }
+
+ glsl_to_tgsi_instruction *inst;
+
+ if (!strcmp("__intrinsic_load_shared", callee)) {
+ inst = emit_asm(ir, TGSI_OPCODE_LOAD, dst, off);
+ inst->buffer = buffer;
+ } else if (!strcmp("__intrinsic_store_shared", callee)) {
+ param = param->get_next();
+ ir_rvalue *val = ((ir_instruction *)param)->as_rvalue();
+ val->accept(this);
+
+ param = param->get_next();
+ ir_constant *write_mask = ((ir_instruction *)param)->as_constant();
+ assert(write_mask);
+ dst.writemask = write_mask->value.u[0];
+
+ dst.type = this->result.type;
+ inst = emit_asm(ir, TGSI_OPCODE_STORE, dst, off, this->result);
+ inst->buffer = buffer;
+ } else {
+ param = param->get_next();
+ ir_rvalue *val = ((ir_instruction *)param)->as_rvalue();
+ val->accept(this);
+
+ st_src_reg data = this->result, data2 = undef_src;
+ unsigned opcode;
+ if (!strcmp("__intrinsic_atomic_add_shared", callee))
+ opcode = TGSI_OPCODE_ATOMUADD;
+ else if (!strcmp("__intrinsic_atomic_min_shared", callee))
+ opcode = TGSI_OPCODE_ATOMIMIN;
+ else if (!strcmp("__intrinsic_atomic_max_shared", callee))
+ opcode = TGSI_OPCODE_ATOMIMAX;
+ else if (!strcmp("__intrinsic_atomic_and_shared", callee))
+ opcode = TGSI_OPCODE_ATOMAND;
+ else if (!strcmp("__intrinsic_atomic_or_shared", callee))
+ opcode = TGSI_OPCODE_ATOMOR;
+ else if (!strcmp("__intrinsic_atomic_xor_shared", callee))
+ opcode = TGSI_OPCODE_ATOMXOR;
+ else if (!strcmp("__intrinsic_atomic_exchange_shared", callee))
+ opcode = TGSI_OPCODE_ATOMXCHG;
+ else if (!strcmp("__intrinsic_atomic_comp_swap_shared", callee)) {
+ opcode = TGSI_OPCODE_ATOMCAS;
+ param = param->get_next();
+ val = ((ir_instruction *)param)->as_rvalue();
+ val->accept(this);
+ data2 = this->result;
+ } else {
+ assert(!"Unexpected intrinsic");
+ return;
+ }
+
+ inst = emit_asm(ir, opcode, dst, off, data, data2);
+ inst->buffer = buffer;
+ }
+}
+
+void
+glsl_to_tgsi_visitor::visit_image_intrinsic(ir_call *ir)
+{
+ const char *callee = ir->callee->function_name();
+ exec_node *param = ir->actual_parameters.get_head();
+
+ ir_dereference *img = (ir_dereference *)param;
+ const ir_variable *imgvar = img->variable_referenced();
+ const glsl_type *type = imgvar->type->without_array();
+ unsigned sampler_array_size = 1, sampler_base = 0;
+
+ st_src_reg reladdr;
+ st_src_reg image(PROGRAM_IMAGE, 0, GLSL_TYPE_UINT);
+
+ get_deref_offsets(img, &sampler_array_size, &sampler_base,
+ (unsigned int *)&image.index, &reladdr);
+ if (reladdr.file != PROGRAM_UNDEFINED) {
+ emit_arl(ir, sampler_reladdr, reladdr);
+ image.reladdr = ralloc(mem_ctx, st_src_reg);
+ memcpy(image.reladdr, &sampler_reladdr, sizeof(reladdr));
+ }
+
+ st_dst_reg dst = undef_dst;
+ if (ir->return_deref) {
+ ir->return_deref->accept(this);
+ dst = st_dst_reg(this->result);
+ dst.writemask = (1 << ir->return_deref->type->vector_elements) - 1;
+ }
+
+ glsl_to_tgsi_instruction *inst;
+
+ if (!strcmp("__intrinsic_image_size", callee)) {
+ dst.writemask = WRITEMASK_XYZ;
+ inst = emit_asm(ir, TGSI_OPCODE_RESQ, dst);
+ } else if (!strcmp("__intrinsic_image_samples", callee)) {
+ st_src_reg res = get_temp(glsl_type::ivec4_type);
+ st_dst_reg dstres = st_dst_reg(res);
+ dstres.writemask = WRITEMASK_W;
+ emit_asm(ir, TGSI_OPCODE_RESQ, dstres);
+ res.swizzle = SWIZZLE_WWWW;
+ inst = emit_asm(ir, TGSI_OPCODE_MOV, dst, res);
+ } else {
+ st_src_reg arg1 = undef_src, arg2 = undef_src;
+ st_src_reg coord;
+ st_dst_reg coord_dst;
+ coord = get_temp(glsl_type::ivec4_type);
+ coord_dst = st_dst_reg(coord);
+ coord_dst.writemask = (1 << type->coordinate_components()) - 1;
+ param = param->get_next();
+ ((ir_dereference *)param)->accept(this);
+ emit_asm(ir, TGSI_OPCODE_MOV, coord_dst, this->result);
+ coord.swizzle = SWIZZLE_XXXX;
+ switch (type->coordinate_components()) {
+ case 4: assert(!"unexpected coord count");
+ /* fallthrough */
+ case 3: coord.swizzle |= SWIZZLE_Z << 6;
+ /* fallthrough */
+ case 2: coord.swizzle |= SWIZZLE_Y << 3;
+ }
+
+ if (type->sampler_dimensionality == GLSL_SAMPLER_DIM_MS) {
+ param = param->get_next();
+ ((ir_dereference *)param)->accept(this);
+ st_src_reg sample = this->result;
+ sample.swizzle = SWIZZLE_XXXX;
+ coord_dst.writemask = WRITEMASK_W;
+ emit_asm(ir, TGSI_OPCODE_MOV, coord_dst, sample);
+ coord.swizzle |= SWIZZLE_W << 9;
+ }
+
+ param = param->get_next();
+ if (!param->is_tail_sentinel()) {
+ ((ir_dereference *)param)->accept(this);
+ arg1 = this->result;
+ param = param->get_next();
+ }
+
+ if (!param->is_tail_sentinel()) {
+ ((ir_dereference *)param)->accept(this);
+ arg2 = this->result;
+ param = param->get_next();
+ }
+
+ assert(param->is_tail_sentinel());
+
+ unsigned opcode;
+ if (!strcmp("__intrinsic_image_load", callee))
+ opcode = TGSI_OPCODE_LOAD;
+ else if (!strcmp("__intrinsic_image_store", callee))
+ opcode = TGSI_OPCODE_STORE;
+ else if (!strcmp("__intrinsic_image_atomic_add", callee))
+ opcode = TGSI_OPCODE_ATOMUADD;
+ else if (!strcmp("__intrinsic_image_atomic_min", callee))
+ opcode = TGSI_OPCODE_ATOMIMIN;
+ else if (!strcmp("__intrinsic_image_atomic_max", callee))
+ opcode = TGSI_OPCODE_ATOMIMAX;
+ else if (!strcmp("__intrinsic_image_atomic_and", callee))
+ opcode = TGSI_OPCODE_ATOMAND;
+ else if (!strcmp("__intrinsic_image_atomic_or", callee))
+ opcode = TGSI_OPCODE_ATOMOR;
+ else if (!strcmp("__intrinsic_image_atomic_xor", callee))
+ opcode = TGSI_OPCODE_ATOMXOR;
+ else if (!strcmp("__intrinsic_image_atomic_exchange", callee))
+ opcode = TGSI_OPCODE_ATOMXCHG;
+ else if (!strcmp("__intrinsic_image_atomic_comp_swap", callee))
+ opcode = TGSI_OPCODE_ATOMCAS;
+ else {
+ assert(!"Unexpected intrinsic");
+ return;
+ }
+
+ inst = emit_asm(ir, opcode, dst, coord, arg1, arg2);
+ if (opcode == TGSI_OPCODE_STORE)
+ inst->dst[0].writemask = WRITEMASK_XYZW;
+ }
+
+ inst->buffer = image;
+ inst->sampler_array_size = sampler_array_size;
+ inst->sampler_base = sampler_base;
+
+ switch (type->sampler_dimensionality) {
+ case GLSL_SAMPLER_DIM_1D:
+ inst->tex_target = (type->sampler_array)
+ ? TEXTURE_1D_ARRAY_INDEX : TEXTURE_1D_INDEX;
+ break;
+ case GLSL_SAMPLER_DIM_2D:
+ inst->tex_target = (type->sampler_array)
+ ? TEXTURE_2D_ARRAY_INDEX : TEXTURE_2D_INDEX;
+ break;
+ case GLSL_SAMPLER_DIM_3D:
+ inst->tex_target = TEXTURE_3D_INDEX;
+ break;
+ case GLSL_SAMPLER_DIM_CUBE:
+ inst->tex_target = (type->sampler_array)
+ ? TEXTURE_CUBE_ARRAY_INDEX : TEXTURE_CUBE_INDEX;
+ break;
+ case GLSL_SAMPLER_DIM_RECT:
+ inst->tex_target = TEXTURE_RECT_INDEX;
+ break;
+ case GLSL_SAMPLER_DIM_BUF:
+ inst->tex_target = TEXTURE_BUFFER_INDEX;
+ break;
+ case GLSL_SAMPLER_DIM_EXTERNAL:
+ inst->tex_target = TEXTURE_EXTERNAL_INDEX;
+ break;
+ case GLSL_SAMPLER_DIM_MS:
+ inst->tex_target = (type->sampler_array)
+ ? TEXTURE_2D_MULTISAMPLE_ARRAY_INDEX : TEXTURE_2D_MULTISAMPLE_INDEX;
+ break;
+ default:
+ assert(!"Should not get here.");
+ }
+
+ 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)
+ inst->buffer_access |= TGSI_MEMORY_COHERENT;
+ if (imgvar->data.image_restrict)
+ inst->buffer_access |= TGSI_MEMORY_RESTRICT;
+ if (imgvar->data.image_volatile)
+ inst->buffer_access |= TGSI_MEMORY_VOLATILE;
+}
+
void
glsl_to_tgsi_visitor::visit(ir_call *ir)
{
/* Filter out intrinsics */
if (!strcmp("__intrinsic_atomic_read", callee) ||
!strcmp("__intrinsic_atomic_increment", callee) ||
- !strcmp("__intrinsic_atomic_predecrement", callee)) {
+ !strcmp("__intrinsic_atomic_predecrement", callee) ||
+ !strcmp("__intrinsic_atomic_add", callee) ||
+ !strcmp("__intrinsic_atomic_sub", callee) ||
+ !strcmp("__intrinsic_atomic_min", callee) ||
+ !strcmp("__intrinsic_atomic_max", callee) ||
+ !strcmp("__intrinsic_atomic_and", callee) ||
+ !strcmp("__intrinsic_atomic_or", callee) ||
+ !strcmp("__intrinsic_atomic_xor", callee) ||
+ !strcmp("__intrinsic_atomic_exchange", callee) ||
+ !strcmp("__intrinsic_atomic_comp_swap", callee)) {
visit_atomic_counter_intrinsic(ir);
return;
}
return;
}
+ if (!strcmp("__intrinsic_load_shared", callee) ||
+ !strcmp("__intrinsic_store_shared", callee) ||
+ !strcmp("__intrinsic_atomic_add_shared", callee) ||
+ !strcmp("__intrinsic_atomic_min_shared", callee) ||
+ !strcmp("__intrinsic_atomic_max_shared", callee) ||
+ !strcmp("__intrinsic_atomic_and_shared", callee) ||
+ !strcmp("__intrinsic_atomic_or_shared", callee) ||
+ !strcmp("__intrinsic_atomic_xor_shared", callee) ||
+ !strcmp("__intrinsic_atomic_exchange_shared", callee) ||
+ !strcmp("__intrinsic_atomic_comp_swap_shared", callee)) {
+ visit_shared_intrinsic(ir);
+ return;
+ }
+
+ if (!strcmp("__intrinsic_image_load", callee) ||
+ !strcmp("__intrinsic_image_store", callee) ||
+ !strcmp("__intrinsic_image_atomic_add", callee) ||
+ !strcmp("__intrinsic_image_atomic_min", callee) ||
+ !strcmp("__intrinsic_image_atomic_max", callee) ||
+ !strcmp("__intrinsic_image_atomic_and", callee) ||
+ !strcmp("__intrinsic_image_atomic_or", callee) ||
+ !strcmp("__intrinsic_image_atomic_xor", callee) ||
+ !strcmp("__intrinsic_image_atomic_exchange", callee) ||
+ !strcmp("__intrinsic_image_atomic_comp_swap", callee) ||
+ !strcmp("__intrinsic_image_size", callee) ||
+ !strcmp("__intrinsic_image_samples", callee)) {
+ visit_image_intrinsic(ir);
+ return;
+ }
+
entry = get_function_signature(sig);
/* Process in parameters. */
foreach_two_lists(formal_node, &sig->parameters,
l.index = storage->index;
l.reladdr = NULL;
l.writemask = WRITEMASK_XYZW;
- l.cond_mask = COND_TR;
for (i = 0; i < type_size(param->type); i++) {
emit_asm(ir, TGSI_OPCODE_MOV, l, r);
this->result = entry->return_reg;
}
+void
+glsl_to_tgsi_visitor::calc_deref_offsets(ir_dereference *head,
+ ir_dereference *tail,
+ unsigned *array_elements,
+ unsigned *base,
+ unsigned *index,
+ st_src_reg *indirect,
+ unsigned *location)
+{
+ switch (tail->ir_type) {
+ case ir_type_dereference_record: {
+ ir_dereference_record *deref_record = tail->as_dereference_record();
+ const glsl_type *struct_type = deref_record->record->type;
+ int field_index = deref_record->record->type->field_index(deref_record->field);
+
+ calc_deref_offsets(head, deref_record->record->as_dereference(), array_elements, base, index, indirect, location);
+
+ assert(field_index >= 0);
+ *location += struct_type->record_location_offset(field_index);
+ break;
+ }
+
+ case ir_type_dereference_array: {
+ ir_dereference_array *deref_arr = tail->as_dereference_array();
+ ir_constant *array_index = deref_arr->array_index->constant_expression_value();
+
+ if (!array_index) {
+ st_src_reg temp_reg;
+ st_dst_reg temp_dst;
+
+ temp_reg = get_temp(glsl_type::uint_type);
+ temp_dst = st_dst_reg(temp_reg);
+ temp_dst.writemask = 1;
+
+ deref_arr->array_index->accept(this);
+ if (*array_elements != 1)
+ emit_asm(NULL, TGSI_OPCODE_MUL, temp_dst, this->result, st_src_reg_for_int(*array_elements));
+ else
+ emit_asm(NULL, TGSI_OPCODE_MOV, temp_dst, this->result);
+
+ if (indirect->file == PROGRAM_UNDEFINED)
+ *indirect = temp_reg;
+ else {
+ temp_dst = st_dst_reg(*indirect);
+ temp_dst.writemask = 1;
+ emit_asm(NULL, TGSI_OPCODE_ADD, temp_dst, *indirect, temp_reg);
+ }
+ } else
+ *index += array_index->value.u[0] * *array_elements;
+
+ *array_elements *= deref_arr->array->type->length;
+
+ calc_deref_offsets(head, deref_arr->array->as_dereference(), array_elements, base, index, indirect, location);
+ break;
+ }
+ default:
+ break;
+ }
+}
+
+void
+glsl_to_tgsi_visitor::get_deref_offsets(ir_dereference *ir,
+ unsigned *array_size,
+ unsigned *base,
+ unsigned *index,
+ st_src_reg *reladdr)
+{
+ GLuint shader = _mesa_program_enum_to_shader_stage(this->prog->Target);
+ unsigned location = 0;
+ ir_variable *var = ir->variable_referenced();
+
+ memset(reladdr, 0, sizeof(*reladdr));
+ reladdr->file = PROGRAM_UNDEFINED;
+
+ *base = 0;
+ *array_size = 1;
+
+ assert(var);
+ location = var->data.location;
+ calc_deref_offsets(ir, ir, array_size, base, index, reladdr, &location);
+
+ /*
+ * If we end up with no indirect then adjust the base to the index,
+ * and set the array size to 1.
+ */
+ if (reladdr->file == PROGRAM_UNDEFINED) {
+ *base = *index;
+ *array_size = 1;
+ }
+
+ if (location != 0xffffffff) {
+ *base += this->shader_program->UniformStorage[location].opaque[shader].index;
+ *index += this->shader_program->UniformStorage[location].opaque[shader].index;
+ }
+}
+
void
glsl_to_tgsi_visitor::visit(ir_texture *ir)
{
st_src_reg result_src, coord, cube_sc, lod_info, projector, dx, dy;
st_src_reg offset[MAX_GLSL_TEXTURE_OFFSET], sample_index, component;
- st_src_reg levels_src;
+ st_src_reg levels_src, reladdr;
st_dst_reg result_dst, coord_dst, cube_sc_dst;
glsl_to_tgsi_instruction *inst = NULL;
unsigned opcode = TGSI_OPCODE_NOP;
const glsl_type *sampler_type = ir->sampler->type;
- ir_rvalue *sampler_index =
- _mesa_get_sampler_array_nonconst_index(ir->sampler);
+ unsigned sampler_array_size = 1, sampler_index = 0, sampler_base = 0;
bool is_cube_array = false;
unsigned i;
coord_dst.writemask = WRITEMASK_XYZW;
}
- if (sampler_index) {
- sampler_index->accept(this);
- emit_arl(ir, sampler_reladdr, this->result);
- }
+ get_deref_offsets(ir->sampler, &sampler_array_size, &sampler_base,
+ &sampler_index, &reladdr);
+ if (reladdr.file != PROGRAM_UNDEFINED)
+ emit_arl(ir, sampler_reladdr, reladdr);
if (opcode == TGSI_OPCODE_TXD)
inst = emit_asm(ir, opcode, result_dst, coord, dx, dy);
if (ir->shadow_comparitor)
inst->tex_shadow = GL_TRUE;
- inst->sampler.index = _mesa_get_sampler_uniform_value(ir->sampler,
- this->shader_program,
- this->prog);
- if (sampler_index) {
+ inst->sampler.index = sampler_index;
+ inst->sampler_array_size = sampler_array_size;
+ inst->sampler_base = sampler_base;
+
+ if (reladdr.file != PROGRAM_UNDEFINED) {
inst->sampler.reladdr = ralloc(mem_ctx, st_src_reg);
- memcpy(inst->sampler.reladdr, &sampler_reladdr, sizeof(sampler_reladdr));
- inst->sampler_array_size =
- ir->sampler->as_dereference_array()->array->type->array_size();
- } else {
- inst->sampler_array_size = 1;
+ memcpy(inst->sampler.reladdr, &reladdr, sizeof(reladdr));
}
if (ir->offset) {
num_address_regs = 0;
samplers_used = 0;
buffers_used = 0;
+ images_used = 0;
indirect_addr_consts = false;
wpos_transform_const = -1;
glsl_version = 0;
options = NULL;
have_sqrt = false;
have_fma = false;
+ use_shared_memory = false;
}
glsl_to_tgsi_visitor::~glsl_to_tgsi_visitor()
{
v->samplers_used = 0;
v->buffers_used = 0;
+ v->images_used = 0;
foreach_in_list(glsl_to_tgsi_instruction, inst, &v->instructions) {
if (inst->info->is_tex) {
for (int i = 0; i < inst->sampler_array_size; i++) {
- unsigned idx = inst->sampler.index + i;
+ unsigned idx = inst->sampler_base + i;
v->samplers_used |= 1 << idx;
debug_assert(idx < (int)ARRAY_SIZE(v->sampler_types));
if (inst->buffer.file != PROGRAM_UNDEFINED && (
is_resource_instruction(inst->op) ||
inst->op == TGSI_OPCODE_STORE)) {
- if (inst->buffer.file == PROGRAM_BUFFER)
+ if (inst->buffer.file == PROGRAM_BUFFER) {
v->buffers_used |= 1 << inst->buffer.index;
+ } else if (inst->buffer.file == PROGRAM_MEMORY) {
+ v->use_shared_memory = true;
+ } else {
+ assert(inst->buffer.file == PROGRAM_IMAGE);
+ for (int i = 0; i < inst->sampler_array_size; i++) {
+ unsigned idx = inst->sampler_base + i;
+ v->images_used |= 1 << idx;
+ v->image_targets[idx] =
+ st_translate_texture_target(inst->tex_target, false);
+ v->image_formats[idx] = inst->image_format;
+ }
+ }
}
}
prog->SamplersUsed = v->samplers_used;
struct ureg_dst address[3];
struct ureg_src samplers[PIPE_MAX_SAMPLERS];
struct ureg_src buffers[PIPE_MAX_SHADER_BUFFERS];
+ struct ureg_src images[PIPE_MAX_SHADER_IMAGES];
struct ureg_src systemValues[SYSTEM_VALUE_MAX];
+ struct ureg_src shared_memory;
struct tgsi_texture_offset tex_offsets[MAX_GLSL_TEXTURE_OFFSET];
unsigned *array_sizes;
struct array_decl *input_arrays;
TGSI_SEMANTIC_PRIMID,
TGSI_SEMANTIC_TESSOUTER,
TGSI_SEMANTIC_TESSINNER,
+
+ /* Compute shaders
+ */
+ TGSI_SEMANTIC_THREAD_ID,
+ TGSI_SEMANTIC_BLOCK_ID,
+ TGSI_SEMANTIC_GRID_SIZE,
};
/**
int num_dst;
int num_src;
- unsigned tex_target;
+ unsigned tex_target = 0;
num_dst = num_inst_dst_regs(inst);
num_src = num_inst_src_regs(inst);
for (i = num_src - 1; i >= 0; i--)
src[i + 1] = src[i];
num_src++;
- src[0] = t->buffers[inst->buffer.index];
+ if (inst->buffer.file == PROGRAM_MEMORY) {
+ src[0] = t->shared_memory;
+ } else if (inst->buffer.file == PROGRAM_BUFFER) {
+ src[0] = t->buffers[inst->buffer.index];
+ } else {
+ src[0] = t->images[inst->buffer.index];
+ tex_target = st_translate_texture_target(inst->tex_target, inst->tex_shadow);
+ }
if (inst->buffer.reladdr)
src[0] = ureg_src_indirect(src[0], ureg_src(t->address[2]));
assert(src[0].File != TGSI_FILE_NULL);
ureg_memory_insn(ureg, inst->op, dst, num_dst, src, num_src,
- inst->buffer_access);
+ inst->buffer_access,
+ tex_target, inst->image_format);
break;
case TGSI_OPCODE_STORE:
- dst[0] = ureg_writemask(ureg_dst(t->buffers[inst->buffer.index]), inst->dst[0].writemask);
+ if (inst->buffer.file == PROGRAM_MEMORY) {
+ dst[0] = ureg_dst(t->shared_memory);
+ } else if (inst->buffer.file == PROGRAM_BUFFER) {
+ dst[0] = ureg_dst(t->buffers[inst->buffer.index]);
+ } else {
+ dst[0] = ureg_dst(t->images[inst->buffer.index]);
+ tex_target = st_translate_texture_target(inst->tex_target, inst->tex_shadow);
+ }
+ dst[0] = ureg_writemask(dst[0], inst->dst[0].writemask);
if (inst->buffer.reladdr)
dst[0] = ureg_dst_indirect(dst[0], ureg_src(t->address[2]));
assert(dst[0].File != TGSI_FILE_NULL);
ureg_memory_insn(ureg, inst->op, dst, num_dst, src, num_src,
- inst->buffer_access);
+ inst->buffer_access,
+ tex_target, inst->image_format);
break;
case TGSI_OPCODE_SCS:
TGSI_SEMANTIC_TESSCOORD);
assert(_mesa_sysval_to_semantic[SYSTEM_VALUE_HELPER_INVOCATION] ==
TGSI_SEMANTIC_HELPER_INVOCATION);
+ assert(_mesa_sysval_to_semantic[SYSTEM_VALUE_LOCAL_INVOCATION_ID] ==
+ TGSI_SEMANTIC_THREAD_ID);
+ assert(_mesa_sysval_to_semantic[SYSTEM_VALUE_WORK_GROUP_ID] ==
+ TGSI_SEMANTIC_BLOCK_ID);
+ assert(_mesa_sysval_to_semantic[SYSTEM_VALUE_NUM_WORK_GROUPS] ==
+ TGSI_SEMANTIC_GRID_SIZE);
t = CALLOC_STRUCT(st_translate);
if (!t) {
t->inputs[i] = ureg_DECL_vs_input(ureg, i);
}
break;
+ case TGSI_PROCESSOR_COMPUTE:
+ break;
default:
assert(0);
}
*/
switch (procType) {
case TGSI_PROCESSOR_FRAGMENT:
+ case TGSI_PROCESSOR_COMPUTE:
break;
case TGSI_PROCESSOR_GEOMETRY:
case TGSI_PROCESSOR_TESS_EVAL:
}
if (procType == TGSI_PROCESSOR_FRAGMENT) {
+ if (program->shader->EarlyFragmentTests)
+ ureg_property(ureg, TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL, 1);
+
if (proginfo->InputsRead & VARYING_BIT_POS) {
/* Must do this after setting up t->inputs. */
emit_wpos(st_context(ctx), t, proginfo, ureg,
}
}
+ if (program->use_shared_memory)
+ t->shared_memory = ureg_DECL_memory(ureg, TGSI_MEMORY_TYPE_SHARED);
+ for (i = 0; i < program->shader->NumImages; i++) {
+ if (program->images_used & (1 << i)) {
+ t->images[i] = ureg_DECL_image(ureg, i,
+ program->image_targets[i],
+ program->image_formats[i],
+ true, false);
+ }
+ }
/* Emit each instruction in turn:
*/
t->insn[t->labels[i].branch_target]);
}
+ /* Set the next shader stage hint for VS and TES. */
+ switch (procType) {
+ case TGSI_PROCESSOR_VERTEX:
+ case TGSI_PROCESSOR_TESS_EVAL:
+ if (program->shader_program->SeparateShader)
+ break;
+
+ for (i = program->shader->Stage+1; i <= MESA_SHADER_FRAGMENT; i++) {
+ if (program->shader_program->_LinkedShaders[i]) {
+ unsigned next;
+
+ switch (i) {
+ case MESA_SHADER_TESS_CTRL:
+ next = TGSI_PROCESSOR_TESS_CTRL;
+ break;
+ case MESA_SHADER_TESS_EVAL:
+ next = TGSI_PROCESSOR_TESS_EVAL;
+ break;
+ case MESA_SHADER_GEOMETRY:
+ next = TGSI_PROCESSOR_GEOMETRY;
+ break;
+ case MESA_SHADER_FRAGMENT:
+ next = TGSI_PROCESSOR_FRAGMENT;
+ break;
+ default:
+ assert(0);
+ continue;
+ }
+
+ ureg_set_next_shader_processor(ureg, next);
+ break;
+ }
+ }
+ break;
+ }
+
out:
if (t) {
free(t->arrays);
prog->OutputsWritten, 0ULL, prog->PatchOutputsWritten);
count_resources(v, prog);
+ /* The GLSL IR won't be needed anymore. */
+ ralloc_free(shader->ir);
+ shader->ir = NULL;
+
/* This must be done before the uniform storage is associated. */
if (shader->Type == GL_FRAGMENT_SHADER &&
(prog->InputsRead & VARYING_BIT_POS ||
struct st_geometry_program *stgp;
struct st_tessctrl_program *sttcp;
struct st_tesseval_program *sttep;
+ struct st_compute_program *stcp;
switch (shader->Type) {
case GL_VERTEX_SHADER:
sttep = (struct st_tesseval_program *)prog;
sttep->glsl_to_tgsi = v;
break;
+ case GL_COMPUTE_SHADER:
+ stcp = (struct st_compute_program *)prog;
+ stcp->glsl_to_tgsi = v;
+ break;
default:
assert(!"should not be reached");
return NULL;
validate_ir_tree(ir);
}
+ build_program_resource_list(prog);
+
for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
struct gl_program *linked_prog;