}
} else if (ir->location == FRAG_RESULT_DEPTH) {
this->frag_depth = *reg;
+ } else if (ir->location == FRAG_RESULT_SAMPLE_MASK) {
+ this->sample_mask = *reg;
} else {
/* gl_FragData or a user-defined FS output */
assert(ir->location >= FRAG_RESULT_DATA0 &&
reg = new(this->mem_ctx) fs_reg(UNIFORM, param_index);
reg->type = brw_type_for_base_type(ir->type);
+
+ } else if (ir->mode == ir_var_system_value) {
+ if (ir->location == SYSTEM_VALUE_SAMPLE_POS) {
+ reg = emit_samplepos_setup(ir);
+ } else if (ir->location == SYSTEM_VALUE_SAMPLE_ID) {
+ reg = emit_sampleid_setup(ir);
+ }
}
if (!reg)
break;
case GLSL_TYPE_SAMPLER:
+ case GLSL_TYPE_ATOMIC_UINT:
break;
case GLSL_TYPE_VOID:
const int vector_elements =
ir->coordinate ? ir->coordinate->type->vector_elements : 0;
- if (ir->offset != NULL && ir->op == ir_txf) {
- /* It appears that the ld instruction used for txf does its
- * address bounds check before adding in the offset. To work
- * around this, just add the integer offset to the integer texel
- * coordinate, and don't put the offset in the header.
+ if (ir->offset) {
+ /* The offsets set up by the ir_texture visitor are in the
+ * m1 header, so we can't go headerless.
*/
- ir_constant *offset = ir->offset->as_constant();
- for (int i = 0; i < vector_elements; i++) {
- emit(ADD(fs_reg(MRF, base_mrf + mlen + i * reg_width, coordinate.type),
- coordinate,
- offset->value.i[i]));
- coordinate.reg_offset++;
- }
- } else {
- if (ir->offset) {
- /* The offsets set up by the ir_texture visitor are in the
- * m1 header, so we can't go headerless.
- */
- header_present = true;
- mlen++;
- base_mrf--;
- }
+ header_present = true;
+ mlen++;
+ base_mrf--;
+ }
- for (int i = 0; i < vector_elements; i++) {
- emit(MOV(fs_reg(MRF, base_mrf + mlen + i * reg_width, coordinate.type),
- coordinate));
- coordinate.reg_offset++;
- }
+ for (int i = 0; i < vector_elements; i++) {
+ emit(MOV(fs_reg(MRF, base_mrf + mlen + i * reg_width, coordinate.type),
+ coordinate));
+ coordinate.reg_offset++;
}
mlen += vector_elements * reg_width;
{
int reg_width = dispatch_width / 8;
bool header_present = false;
- int offsets[3];
fs_reg payload = fs_reg(this, glsl_type::float_type);
fs_reg next = payload;
next.reg_offset++;
}
+ bool has_nonconstant_offset = ir->offset && !ir->offset->as_constant();
+ bool coordinate_done = false;
+
/* Set up the LOD info */
switch (ir->op) {
case ir_tex:
case ir_lod:
- case ir_tg4:
break;
case ir_txb:
emit(MOV(next, lod));
next.reg_offset++;
}
}
+
+ coordinate_done = true;
break;
}
case ir_txs:
next.reg_offset++;
break;
case ir_txf:
- /* It appears that the ld instruction used for txf does its
- * address bounds check before adding in the offset. To work
- * around this, just add the integer offset to the integer texel
- * coordinate, and don't put the offset in the header.
- */
- if (ir->offset) {
- ir_constant *offset = ir->offset->as_constant();
- offsets[0] = offset->value.i[0];
- offsets[1] = offset->value.i[1];
- offsets[2] = offset->value.i[2];
- } else {
- memset(offsets, 0, sizeof(offsets));
- }
-
/* Unfortunately, the parameters for LD are intermixed: u, lod, v, r. */
- emit(ADD(next.retype(BRW_REGISTER_TYPE_D), coordinate, offsets[0]));
+ emit(MOV(next.retype(BRW_REGISTER_TYPE_D), coordinate));
coordinate.reg_offset++;
next.reg_offset++;
next.reg_offset++;
for (int i = 1; i < ir->coordinate->type->vector_elements; i++) {
- emit(ADD(next.retype(BRW_REGISTER_TYPE_D), coordinate, offsets[i]));
+ emit(MOV(next.retype(BRW_REGISTER_TYPE_D), coordinate));
coordinate.reg_offset++;
next.reg_offset++;
}
+
+ coordinate_done = true;
break;
case ir_txf_ms:
emit(MOV(next.retype(BRW_REGISTER_TYPE_UD), sample_index));
coordinate.reg_offset++;
next.reg_offset++;
}
+
+ coordinate_done = true;
+ break;
+ case ir_tg4:
+ if (has_nonconstant_offset) {
+ if (ir->shadow_comparitor && dispatch_width == 16)
+ fail("Gen7 does not support gather4_po_c in SIMD16 mode.");
+
+ /* More crazy intermixing */
+ ir->offset->accept(this);
+ fs_reg offset_value = this->result;
+
+ for (int i = 0; i < 2; i++) { /* u, v */
+ emit(MOV(next, coordinate));
+ coordinate.reg_offset++;
+ next.reg_offset++;
+ }
+
+ for (int i = 0; i < 2; i++) { /* offu, offv */
+ emit(MOV(next.retype(BRW_REGISTER_TYPE_D), offset_value));
+ offset_value.reg_offset++;
+ next.reg_offset++;
+ }
+
+ if (ir->coordinate->type->vector_elements == 3) { /* r if present */
+ emit(MOV(next, coordinate));
+ coordinate.reg_offset++;
+ next.reg_offset++;
+ }
+
+ coordinate_done = true;
+ }
break;
}
/* Set up the coordinate (except for cases where it was done above) */
- if (ir->op != ir_txd && ir->op != ir_txs && ir->op != ir_txf && ir->op != ir_txf_ms && ir->op != ir_query_levels) {
+ if (ir->coordinate && !coordinate_done) {
for (int i = 0; i < ir->coordinate->type->vector_elements; i++) {
- emit(MOV(next, coordinate));
- coordinate.reg_offset++;
- next.reg_offset++;
+ emit(MOV(next, coordinate));
+ coordinate.reg_offset++;
+ next.reg_offset++;
}
}
case ir_txs: inst = emit(SHADER_OPCODE_TXS, dst, payload); break;
case ir_query_levels: inst = emit(SHADER_OPCODE_TXS, dst, payload); break;
case ir_lod: inst = emit(SHADER_OPCODE_LOD, dst, payload); break;
- case ir_tg4: inst = emit(SHADER_OPCODE_TG4, dst, payload); break;
+ case ir_tg4:
+ if (has_nonconstant_offset)
+ inst = emit(SHADER_OPCODE_TG4_OFFSET, dst, payload);
+ else
+ inst = emit(SHADER_OPCODE_TG4, dst, payload);
+ break;
}
inst->base_mrf = -1;
if (reg_width == 2)
inst->mlen = next.reg_offset * reg_width - header_present;
else
inst->mlen = next.reg_offset * reg_width;
-
inst->header_present = header_present;
inst->regs_written = 4;
/* Should be lowered by do_lower_texture_projection */
assert(!ir->projector);
+ /* Should be lowered */
+ assert(!ir->offset || !ir->offset->type->is_array());
+
/* Generate code to compute all the subexpression trees. This has to be
* done before loading any values into MRFs for the sampler message since
* generating these values may involve SEND messages that need the MRFs.
}
if (ir->offset != NULL && ir->op != ir_txf)
- inst->texture_offset = brw_texture_offset(ir->offset->as_constant());
+ inst->texture_offset = brw_texture_offset(ctx, ir->offset->as_constant());
if (ir->op == ir_tg4)
inst->texture_offset |= gather_channel(ir, sampler) << 16; // M0.2:16-17
pop_force_uncompressed();
}
+ c->prog_data.uses_omask =
+ fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
+ if(c->prog_data.uses_omask) {
+ this->current_annotation = "FB write oMask";
+ assert(this->sample_mask.file != BAD_FILE);
+ /* Hand over gl_SampleMask. Only lower 16 bits are relevant. */
+ emit(FS_OPCODE_SET_OMASK, fs_reg(MRF, nr, BRW_REGISTER_TYPE_UW), this->sample_mask);
+ nr += 1;
+ }
+
/* Reserve space for color. It'll be filled in per MRT below. */
int color_mrf = nr;
nr += 4 * reg_width;
this->c = c;
this->brw = brw;
this->fp = fp;
+ this->prog = &fp->Base;
this->shader_prog = shader_prog;
this->prog = &fp->Base;
+ this->stage_prog_data = &c->prog_data.base;
this->ctx = &brw->ctx;
this->mem_ctx = ralloc_context(NULL);
if (shader_prog)
this->force_uncompressed_stack = 0;
this->force_sechalf_stack = 0;
+ this->spilled_any_registers = false;
+
memset(&this->param_size, 0, sizeof(this->param_size));
}