#include "brw_cfg.h"
#include "brw_nir.h"
#include "brw_vec4_builder.h"
-#include "brw_vec4_live_variables.h"
#include "brw_vec4_vs.h"
#include "brw_dead_control_flow.h"
-#include "common/gen_debug.h"
+#include "dev/gen_debug.h"
#include "program/prog_parameter.h"
#include "util/u_math.h"
}
bool
-vec4_instruction::is_send_from_grf()
+vec4_instruction::is_send_from_grf() const
{
switch (opcode) {
case SHADER_OPCODE_SHADER_TIME_ADD:
* instruction -- the generate_* functions generate additional MOVs
* for setup.
*/
-int
-vec4_visitor::implied_mrf_writes(vec4_instruction *inst)
+unsigned
+vec4_instruction::implied_mrf_writes() const
{
- if (inst->mlen == 0 || inst->is_send_from_grf())
+ if (mlen == 0 || is_send_from_grf())
return 0;
- switch (inst->opcode) {
+ switch (opcode) {
case SHADER_OPCODE_RCP:
case SHADER_OPCODE_RSQ:
case SHADER_OPCODE_SQRT:
case SHADER_OPCODE_TG4_OFFSET:
case SHADER_OPCODE_SAMPLEINFO:
case SHADER_OPCODE_GET_BUFFER_SIZE:
- return inst->header_size;
+ return header_size;
default:
unreachable("not reached");
}
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTIONS);
return progress;
}
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTION_DETAIL);
return progress;
}
void
vec4_visitor::pack_uniform_registers()
{
+ if (!compiler->compact_params)
+ return;
+
uint8_t chans_used[this->uniforms];
int new_loc[this->uniforms];
int new_chan[this->uniforms];
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTION_DATA_FLOW |
+ DEPENDENCY_INSTRUCTION_DETAIL);
return progress;
}
if (devinfo->gen == 6 && is_math() && swizzle != BRW_SWIZZLE_XYZW)
return false;
+ /* If we write to the flag register changing the swizzle would change
+ * what channels are written to the flag register.
+ */
+ if (writes_flag())
+ return false;
+
/* We can't swizzle implicit accumulator access. We'd have to
* reswizzle the producer of the accumulator value in addition
* to the consumer (i.e. both MUL and MACH). Just skip this.
opcode != BRW_OPCODE_DP3 && opcode != BRW_OPCODE_DP2 &&
opcode != VEC4_OPCODE_PACK_BYTES) {
for (int i = 0; i < 3; i++) {
- if (src[i].file == BAD_FILE || src[i].file == IMM)
+ if (src[i].file == BAD_FILE)
continue;
+ if (src[i].file == IMM) {
+ assert(src[i].type != BRW_REGISTER_TYPE_V &&
+ src[i].type != BRW_REGISTER_TYPE_UV);
+
+ /* Vector immediate types need to be reswizzled. */
+ if (src[i].type == BRW_REGISTER_TYPE_VF) {
+ const unsigned imm[] = {
+ (src[i].ud >> 0) & 0x0ff,
+ (src[i].ud >> 8) & 0x0ff,
+ (src[i].ud >> 16) & 0x0ff,
+ (src[i].ud >> 24) & 0x0ff,
+ };
+
+ src[i] = brw_imm_vf4(imm[BRW_GET_SWZ(swizzle, 0)],
+ imm[BRW_GET_SWZ(swizzle, 1)],
+ imm[BRW_GET_SWZ(swizzle, 2)],
+ imm[BRW_GET_SWZ(swizzle, 3)]);
+ }
+
+ continue;
+ }
+
src[i].swizzle = brw_compose_swizzle(swizzle, src[i].swizzle);
}
}
{
bool progress = false;
int next_ip = 0;
-
- calculate_live_intervals();
+ const vec4_live_variables &live = live_analysis.require();
foreach_block_and_inst_safe (block, vec4_instruction, inst, cfg) {
int ip = next_ip;
/* Can't coalesce this GRF if someone else was going to
* read it later.
*/
- if (var_range_end(var_from_reg(alloc, dst_reg(inst->src[0])), 8) > ip)
+ if (live.var_range_end(var_from_reg(alloc, dst_reg(inst->src[0])), 8) > ip)
continue;
/* We need to check interference with the final destination between this
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTIONS);
return progress;
}
}
}
+ if (progress)
+ invalidate_analysis(DEPENDENCY_INSTRUCTION_DETAIL);
+
return progress;
}
}
}
}
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTION_DETAIL | DEPENDENCY_VARIABLES);
}
void
-vec4_visitor::dump_instruction(backend_instruction *be_inst)
+vec4_visitor::dump_instruction(const backend_instruction *be_inst) const
{
dump_instruction(be_inst, stderr);
}
void
-vec4_visitor::dump_instruction(backend_instruction *be_inst, FILE *file)
+vec4_visitor::dump_instruction(const backend_instruction *be_inst, FILE *file) const
{
- vec4_instruction *inst = (vec4_instruction *)be_inst;
+ const vec4_instruction *inst = (const vec4_instruction *)be_inst;
if (inst->predicate) {
fprintf(file, "(%cf%d.%d%s) ",
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTIONS);
return progress;
}
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTION_DETAIL |
+ DEPENDENCY_VARIABLES);
}
void
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTIONS | DEPENDENCY_VARIABLES);
return progress;
}
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTIONS);
return progress;
}
}
if (progress)
- invalidate_live_intervals();
+ invalidate_analysis(DEPENDENCY_INSTRUCTIONS | DEPENDENCY_VARIABLES);
return progress;
}
}
}
+void
+vec4_visitor::invalidate_analysis(brw::analysis_dependency_class c)
+{
+ backend_shader::invalidate_analysis(c);
+ live_analysis.invalidate(c);
+}
+
bool
vec4_visitor::run()
{
void *mem_ctx,
const struct brw_vs_prog_key *key,
struct brw_vs_prog_data *prog_data,
- nir_shader *shader,
+ nir_shader *nir,
int shader_time_index,
+ struct brw_compile_stats *stats,
char **error_str)
{
const bool is_scalar = compiler->scalar_stage[MESA_SHADER_VERTEX];
- shader = brw_nir_apply_sampler_key(shader, compiler, &key->tex, is_scalar);
+ brw_nir_apply_key(nir, compiler, &key->base, 8, is_scalar);
const unsigned *assembly = NULL;
*/
assert(!is_scalar);
assert(key->copy_edgeflag);
- shader->info.inputs_read |= VERT_BIT_EDGEFLAG;
+ nir->info.inputs_read |= VERT_BIT_EDGEFLAG;
}
- prog_data->inputs_read = shader->info.inputs_read;
- prog_data->double_inputs_read = shader->info.vs.double_inputs;
+ prog_data->inputs_read = nir->info.inputs_read;
+ prog_data->double_inputs_read = nir->info.vs.double_inputs;
- brw_nir_lower_vs_inputs(shader, key->gl_attrib_wa_flags);
- brw_nir_lower_vue_outputs(shader);
- shader = brw_postprocess_nir(shader, compiler, is_scalar);
+ brw_nir_lower_vs_inputs(nir, key->gl_attrib_wa_flags);
+ brw_nir_lower_vue_outputs(nir);
+ brw_postprocess_nir(nir, compiler, is_scalar);
prog_data->base.clip_distance_mask =
- ((1 << shader->info.clip_distance_array_size) - 1);
+ ((1 << nir->info.clip_distance_array_size) - 1);
prog_data->base.cull_distance_mask =
- ((1 << shader->info.cull_distance_array_size) - 1) <<
- shader->info.clip_distance_array_size;
+ ((1 << nir->info.cull_distance_array_size) - 1) <<
+ nir->info.clip_distance_array_size;
unsigned nr_attribute_slots = util_bitcount64(prog_data->inputs_read);
/* gl_VertexID and gl_InstanceID are system values, but arrive via an
* incoming vertex attribute. So, add an extra slot.
*/
- if (shader->info.system_values_read &
+ if (nir->info.system_values_read &
(BITFIELD64_BIT(SYSTEM_VALUE_FIRST_VERTEX) |
BITFIELD64_BIT(SYSTEM_VALUE_BASE_INSTANCE) |
BITFIELD64_BIT(SYSTEM_VALUE_VERTEX_ID_ZERO_BASE) |
}
/* gl_DrawID and IsIndexedDraw share its very own vec4 */
- if (shader->info.system_values_read &
+ if (nir->info.system_values_read &
(BITFIELD64_BIT(SYSTEM_VALUE_DRAW_ID) |
BITFIELD64_BIT(SYSTEM_VALUE_IS_INDEXED_DRAW))) {
nr_attribute_slots++;
}
- if (shader->info.system_values_read &
+ if (nir->info.system_values_read &
BITFIELD64_BIT(SYSTEM_VALUE_IS_INDEXED_DRAW))
prog_data->uses_is_indexed_draw = true;
- if (shader->info.system_values_read &
+ if (nir->info.system_values_read &
BITFIELD64_BIT(SYSTEM_VALUE_FIRST_VERTEX))
prog_data->uses_firstvertex = true;
- if (shader->info.system_values_read &
+ if (nir->info.system_values_read &
BITFIELD64_BIT(SYSTEM_VALUE_BASE_INSTANCE))
prog_data->uses_baseinstance = true;
- if (shader->info.system_values_read &
+ if (nir->info.system_values_read &
BITFIELD64_BIT(SYSTEM_VALUE_VERTEX_ID_ZERO_BASE))
prog_data->uses_vertexid = true;
- if (shader->info.system_values_read &
+ if (nir->info.system_values_read &
BITFIELD64_BIT(SYSTEM_VALUE_INSTANCE_ID))
prog_data->uses_instanceid = true;
- if (shader->info.system_values_read &
+ if (nir->info.system_values_read &
BITFIELD64_BIT(SYSTEM_VALUE_DRAW_ID))
prog_data->uses_drawid = true;
if (is_scalar) {
prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8;
- fs_visitor v(compiler, log_data, mem_ctx, key, &prog_data->base.base,
- NULL, /* prog; Only used for TEXTURE_RECTANGLE on gen < 8 */
- shader, 8, shader_time_index);
+ fs_visitor v(compiler, log_data, mem_ctx, &key->base,
+ &prog_data->base.base,
+ nir, 8, shader_time_index);
if (!v.run_vs()) {
if (error_str)
*error_str = ralloc_strdup(mem_ctx, v.fail_msg);
prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
fs_generator g(compiler, log_data, mem_ctx,
- &prog_data->base.base, v.promoted_constants,
- v.runtime_check_aads_emit, MESA_SHADER_VERTEX);
+ &prog_data->base.base, v.runtime_check_aads_emit,
+ MESA_SHADER_VERTEX);
if (INTEL_DEBUG & DEBUG_VS) {
const char *debug_name =
ralloc_asprintf(mem_ctx, "%s vertex shader %s",
- shader->info.label ? shader->info.label :
+ nir->info.label ? nir->info.label :
"unnamed",
- shader->info.name);
+ nir->info.name);
g.enable_debug(debug_name);
}
- g.generate_code(v.cfg, 8);
+ g.generate_code(v.cfg, 8, v.shader_stats,
+ v.performance_analysis.require(), stats);
+ g.add_const_data(nir->constant_data, nir->constant_data_size);
assembly = g.get_assembly();
}
prog_data->base.dispatch_mode = DISPATCH_MODE_4X2_DUAL_OBJECT;
vec4_vs_visitor v(compiler, log_data, key, prog_data,
- shader, mem_ctx, shader_time_index);
+ nir, mem_ctx, shader_time_index);
if (!v.run()) {
if (error_str)
*error_str = ralloc_strdup(mem_ctx, v.fail_msg);
}
assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx,
- shader, &prog_data->base, v.cfg);
+ nir, &prog_data->base,
+ v.cfg,
+ v.performance_analysis.require(),
+ stats);
}
return assembly;