void *mem_ctx,
const struct brw_wm_prog_key *key,
struct brw_wm_prog_data *prog_data,
- struct gl_shader_program *prog,
+ struct gl_shader_program *shader_prog,
struct gl_fragment_program *fp,
bool runtime_check_aads_emit,
bool debug_flag)
- : brw(brw), key(key), prog_data(prog_data), prog(prog), fp(fp),
- runtime_check_aads_emit(runtime_check_aads_emit), debug_flag(debug_flag),
- mem_ctx(mem_ctx)
+ : brw(brw), stage(MESA_SHADER_FRAGMENT), key(key),
+ prog_data(&prog_data->base), shader_prog(shader_prog),
+ prog(&fp->Base), runtime_check_aads_emit(runtime_check_aads_emit),
+ debug_flag(debug_flag), mem_ctx(mem_ctx)
{
ctx = &brw->ctx;
{
uint32_t msg_control;
+ assert(stage == MESA_SHADER_FRAGMENT);
+ brw_wm_prog_data *prog_data = (brw_wm_prog_data*) this->prog_data;
+
if (brw->gen < 6) {
brw_push_insn_state(p);
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_pop_insn_state(p);
}
- if (prog_data->dual_src_blend)
+ if (inst->opcode == FS_OPCODE_REP_FB_WRITE)
+ msg_control = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE_REPLICATED;
+ else if (prog_data->dual_src_blend)
msg_control = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN01;
else if (dispatch_width == 16)
msg_control = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE;
void
fs_generator::generate_fb_write(fs_inst *inst)
{
+ assert(stage == MESA_SHADER_FRAGMENT);
+ gl_fragment_program *fp = (gl_fragment_program *) prog;
struct brw_reg implied_header;
+ assert(stage == MESA_SHADER_FRAGMENT);
+ brw_wm_prog_data *prog_data = (brw_wm_prog_data*) this->prog_data;
+
/* Header is 2 regs, g0 and g1 are the contents. g0 will be implied
* move, here's g1.
*/
/* On HSW, the GPU will use the predicate on SENDC, unless the header is
* present.
*/
- if ((fp && fp->UsesKill) || key->alpha_test_func) {
+ if (prog_data->uses_kill || key->alpha_test_func) {
struct brw_reg pixel_mask;
if (brw->gen >= 6)
uint32_t base_binding_table_index = (inst->opcode == SHADER_OPCODE_TG4 ||
inst->opcode == SHADER_OPCODE_TG4_OFFSET)
- ? prog_data->base.binding_table.gather_texture_start
- : prog_data->base.binding_table.texture_start;
+ ? prog_data->binding_table.gather_texture_start
+ : prog_data->binding_table.texture_start;
if (sampler_index.file == BRW_IMMEDIATE_VALUE) {
uint32_t sampler = sampler_index.dw1.ud;
simd_mode,
return_format);
- brw_mark_surface_used(&prog_data->base, sampler + base_binding_table_index);
+ brw_mark_surface_used(prog_data, sampler + base_binding_table_index);
} else {
- /* XXX: Non-const sampler index */
+ /* Non-const sampler index */
+ /* Note: this clobbers `dst` as a temporary before emitting the send */
+
+ struct brw_reg addr = vec1(retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD));
+ struct brw_reg temp = vec1(retype(dst, BRW_REGISTER_TYPE_UD));
+
+ struct brw_reg sampler_reg = vec1(retype(sampler_index, BRW_REGISTER_TYPE_UD));
+
+ brw_push_insn_state(p);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+
+ /* Some care required: `sampler` and `temp` may alias:
+ * addr = sampler & 0xff
+ * temp = (sampler << 8) & 0xf00
+ * addr = addr | temp
+ */
+ brw_ADD(p, addr, sampler_reg, brw_imm_ud(base_binding_table_index));
+ brw_SHL(p, temp, sampler_reg, brw_imm_ud(8u));
+ brw_AND(p, temp, temp, brw_imm_ud(0x0f00));
+ brw_AND(p, addr, addr, brw_imm_ud(0x0ff));
+ brw_OR(p, addr, addr, temp);
+
+ /* a0.0 |= <descriptor> */
+ brw_inst *insn_or = brw_next_insn(p, BRW_OPCODE_OR);
+ brw_set_sampler_message(p, insn_or,
+ 0 /* surface */,
+ 0 /* sampler */,
+ msg_type,
+ rlen,
+ inst->mlen /* mlen */,
+ inst->header_present /* header */,
+ simd_mode,
+ return_format);
+ brw_inst_set_exec_size(p->brw, insn_or, BRW_EXECUTE_1);
+ brw_inst_set_src1_reg_type(p->brw, insn_or, BRW_REGISTER_TYPE_UD);
+ brw_set_src0(p, insn_or, addr);
+ brw_set_dest(p, insn_or, addr);
+
+
+ /* dst = send(offset, a0.0) */
+ brw_inst *insn_send = brw_next_insn(p, BRW_OPCODE_SEND);
+ brw_set_dest(p, insn_send, dst);
+ brw_set_src0(p, insn_send, src);
+ brw_set_indirect_send_descriptor(p, insn_send, BRW_SFID_SAMPLER, addr);
+
+ brw_pop_insn_state(p);
+
+ /* visitor knows more than we do about the surface limit required,
+ * so has already done marking.
+ */
}
}
brw_oword_block_read(p, dst, brw_message_reg(inst->base_mrf),
read_offset, surf_index);
- brw_mark_surface_used(&prog_data->base, surf_index);
+ brw_mark_surface_used(prog_data, surf_index);
}
void
BRW_SAMPLER_SIMD_MODE_SIMD4X2,
0);
- brw_mark_surface_used(&prog_data->base, surf_index);
+ brw_mark_surface_used(prog_data, surf_index);
} else {
simd_mode,
return_format);
- brw_mark_surface_used(&prog_data->base, surf_index);
+ brw_mark_surface_used(prog_data, surf_index);
}
void
simd_mode,
0);
- brw_mark_surface_used(&prog_data->base, surf_index);
+ brw_mark_surface_used(prog_data, surf_index);
} else {
brw_MOV(p, payload_offset, offset);
brw_MOV(p, payload_value, value);
brw_shader_time_add(p, payload,
- prog_data->base.binding_table.shader_time_start);
+ prog_data->binding_table.shader_time_start);
brw_pop_insn_state(p);
- brw_mark_surface_used(&prog_data->base,
- prog_data->base.binding_table.shader_time_start);
+ brw_mark_surface_used(prog_data,
+ prog_data->binding_table.shader_time_start);
}
void
atomic_op.dw1.ud, surf_index.dw1.ud,
inst->mlen, dispatch_width / 8);
- brw_mark_surface_used(&prog_data->base, surf_index.dw1.ud);
+ brw_mark_surface_used(prog_data, surf_index.dw1.ud);
}
void
surf_index.dw1.ud,
inst->mlen, dispatch_width / 8);
- brw_mark_surface_used(&prog_data->base, surf_index.dw1.ud);
+ brw_mark_surface_used(prog_data, surf_index.dw1.ud);
}
void
-fs_generator::generate_code(exec_list *instructions)
+fs_generator::generate_code(const cfg_t *cfg)
{
int start_offset = p->next_insn_offset;
+ int loop_count = 0;
struct annotation_info annotation;
memset(&annotation, 0, sizeof(annotation));
- cfg_t *cfg = NULL;
- if (unlikely(debug_flag))
- cfg = new(mem_ctx) cfg_t(instructions);
-
- foreach_in_list(fs_inst, inst, instructions) {
+ foreach_block_and_inst (block, fs_inst, inst, cfg) {
struct brw_reg src[3], dst;
unsigned int last_insn_offset = p->next_insn_offset;
case BRW_OPCODE_WHILE:
brw_WHILE(p);
+ loop_count++;
break;
case SHADER_OPCODE_RCP:
/* Make sure fp->UsesDFdy flag got set (otherwise there's no
* guarantee that key->render_to_fbo is set).
*/
- assert(fp->UsesDFdy);
+ assert(stage == MESA_SHADER_FRAGMENT &&
+ ((gl_fragment_program *) prog)->UsesDFdy);
generate_ddy(inst, dst, src[0], src[1], key->render_to_fbo);
break;
generate_varying_pull_constant_load_gen7(inst, dst, src[0], src[1]);
break;
+ case FS_OPCODE_REP_FB_WRITE:
case FS_OPCODE_FB_WRITE:
generate_fb_write(inst);
break;
int after_size = p->next_insn_offset - start_offset;
if (unlikely(debug_flag)) {
- if (prog) {
+ if (shader_prog) {
fprintf(stderr,
"Native code for %s fragment shader %d (SIMD%d dispatch):\n",
- prog->Label ? prog->Label : "unnamed",
- prog->Name, dispatch_width);
- } else if (fp) {
+ shader_prog->Label ? shader_prog->Label : "unnamed",
+ shader_prog->Name, dispatch_width);
+ } else if (prog) {
fprintf(stderr,
"Native code for fragment program %d (SIMD%d dispatch):\n",
- fp->Base.Id, dispatch_width);
+ prog->Id, dispatch_width);
} else {
fprintf(stderr, "Native code for blorp program (SIMD%d dispatch):\n",
dispatch_width);
}
- fprintf(stderr, "SIMD%d shader: %d instructions. Compacted %d to %d"
+ fprintf(stderr, "SIMD%d shader: %d instructions. %d loops. Compacted %d to %d"
" bytes (%.0f%%)\n",
- dispatch_width, before_size / 16, before_size, after_size,
+ dispatch_width, before_size / 16, loop_count, before_size, after_size,
100.0f * (before_size - after_size) / before_size);
- const struct gl_program *prog = fp ? &fp->Base : NULL;
-
dump_assembly(p->store, annotation.ann_count, annotation.ann, brw, prog);
ralloc_free(annotation.ann);
}
}
const unsigned *
-fs_generator::generate_assembly(exec_list *simd8_instructions,
- exec_list *simd16_instructions,
+fs_generator::generate_assembly(const cfg_t *simd8_cfg,
+ const cfg_t *simd16_cfg,
unsigned *assembly_size)
{
- assert(simd8_instructions || simd16_instructions);
+ assert(simd8_cfg || simd16_cfg);
- if (simd8_instructions) {
+ if (simd8_cfg) {
dispatch_width = 8;
- generate_code(simd8_instructions);
+ generate_code(simd8_cfg);
}
- if (simd16_instructions) {
+ if (simd16_cfg) {
/* align to 64 byte boundary. */
while (p->next_insn_offset % 64) {
brw_NOP(p);
}
+ assert(stage == MESA_SHADER_FRAGMENT);
+ brw_wm_prog_data *prog_data = (brw_wm_prog_data*) this->prog_data;
+
/* Save off the start of this SIMD16 program */
prog_data->prog_offset_16 = p->next_insn_offset;
brw_set_default_compression_control(p, BRW_COMPRESSION_COMPRESSED);
dispatch_width = 16;
- generate_code(simd16_instructions);
+ generate_code(simd16_cfg);
}
return brw_get_program(p, assembly_size);
projects / mesa.git / blobdiff