if (variable_storage(ir))
return;
- if (ir->mode == ir_var_shader_in) {
+ if (ir->data.mode == ir_var_shader_in) {
if (!strcmp(ir->name, "gl_FragCoord")) {
reg = emit_fragcoord_interpolation(ir);
} else if (!strcmp(ir->name, "gl_FrontFacing")) {
assert(reg);
hash_table_insert(this->variable_ht, reg, ir);
return;
- } else if (ir->mode == ir_var_shader_out) {
+ } else if (ir->data.mode == ir_var_shader_out) {
reg = new(this->mem_ctx) fs_reg(this, ir->type);
- if (ir->index > 0) {
- assert(ir->location == FRAG_RESULT_DATA0);
- assert(ir->index == 1);
+ if (ir->data.index > 0) {
+ assert(ir->data.location == FRAG_RESULT_DATA0);
+ assert(ir->data.index == 1);
this->dual_src_output = *reg;
- } else if (ir->location == FRAG_RESULT_COLOR) {
+ } else if (ir->data.location == FRAG_RESULT_COLOR) {
/* Writing gl_FragColor outputs to all color regions. */
for (unsigned int i = 0; i < MAX2(c->key.nr_color_regions, 1); i++) {
this->outputs[i] = *reg;
this->output_components[i] = 4;
}
- } else if (ir->location == FRAG_RESULT_DEPTH) {
+ } else if (ir->data.location == FRAG_RESULT_DEPTH) {
this->frag_depth = *reg;
- } else if (ir->location == FRAG_RESULT_SAMPLE_MASK) {
+ } else if (ir->data.location == FRAG_RESULT_SAMPLE_MASK) {
this->sample_mask = *reg;
} else {
/* gl_FragData or a user-defined FS output */
- assert(ir->location >= FRAG_RESULT_DATA0 &&
- ir->location < FRAG_RESULT_DATA0 + BRW_MAX_DRAW_BUFFERS);
+ assert(ir->data.location >= FRAG_RESULT_DATA0 &&
+ ir->data.location < FRAG_RESULT_DATA0 + BRW_MAX_DRAW_BUFFERS);
int vector_elements =
ir->type->is_array() ? ir->type->fields.array->vector_elements
/* General color output. */
for (unsigned int i = 0; i < MAX2(1, ir->type->length); i++) {
- int output = ir->location - FRAG_RESULT_DATA0 + i;
+ int output = ir->data.location - FRAG_RESULT_DATA0 + i;
this->outputs[output] = *reg;
this->outputs[output].reg_offset += vector_elements * i;
this->output_components[output] = vector_elements;
}
}
- } else if (ir->mode == ir_var_uniform) {
- int param_index = c->prog_data.nr_params;
+ } else if (ir->data.mode == ir_var_uniform) {
+ int param_index = stage_prog_data->nr_params;
/* Thanks to the lower_ubo_reference pass, we will see only
* ir_binop_ubo_load expressions and not ir_dereference_variable for UBO
if (dispatch_width == 16) {
if (!variable_storage(ir)) {
- fail("Failed to find uniform '%s' in 16-wide\n", ir->name);
+ fail("Failed to find uniform '%s' in SIMD16\n", ir->name);
}
return;
}
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) {
+ } else if (ir->data.mode == ir_var_system_value) {
+ if (ir->data.location == SYSTEM_VALUE_SAMPLE_POS) {
reg = emit_samplepos_setup(ir);
- } else if (ir->location == SYSTEM_VALUE_SAMPLE_ID) {
+ } else if (ir->data.location == SYSTEM_VALUE_SAMPLE_ID) {
reg = emit_sampleid_setup(ir);
+ } else if (ir->data.location == SYSTEM_VALUE_SAMPLE_MASK_IN) {
+ reg = emit_samplemaskin_setup(ir);
}
}
* enough.
*/
if (brw->gen >= 7 && dispatch_width == 16)
- fail("16-wide explicit accumulator operands unsupported\n");
+ fail("SIMD16 explicit accumulator operands unsupported\n");
struct brw_reg acc = retype(brw_acc_reg(), this->result.type);
break;
case ir_binop_imul_high: {
if (brw->gen >= 7 && dispatch_width == 16)
- fail("16-wide explicit accumulator operands unsupported\n");
+ fail("SIMD16 explicit accumulator operands unsupported\n");
struct brw_reg acc = retype(brw_acc_reg(), this->result.type);
break;
case ir_binop_carry: {
if (brw->gen >= 7 && dispatch_width == 16)
- fail("16-wide explicit accumulator operands unsupported\n");
+ fail("SIMD16 explicit accumulator operands unsupported\n");
struct brw_reg acc = retype(brw_acc_reg(), BRW_REGISTER_TYPE_UD);
}
case ir_binop_borrow: {
if (brw->gen >= 7 && dispatch_width == 16)
- fail("16-wide explicit accumulator operands unsupported\n");
+ fail("SIMD16 explicit accumulator operands unsupported\n");
struct brw_reg acc = retype(brw_acc_reg(), BRW_REGISTER_TYPE_UD);
emit(fs_inst(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
packed_consts, surf_index, const_offset_reg));
- packed_consts.smear = const_offset->value.u[0] % 16 / 4;
for (int i = 0; i < ir->type->vector_elements; i++) {
+ packed_consts.set_smear(const_offset->value.u[0] % 16 / 4 + i);
+
+ /* The std140 packing rules don't allow vectors to cross 16-byte
+ * boundaries, and a reg is 32 bytes.
+ */
+ assert(packed_consts.subreg_offset < 32);
+
/* UBO bools are any nonzero value. We consider bools to be
* values with the low bit set to 1. Convert them using CMP.
*/
emit(MOV(result, packed_consts));
}
- packed_consts.smear++;
result.reg_offset++;
-
- /* The std140 packing rules don't allow vectors to cross 16-byte
- * boundaries, and a reg is 32 bytes.
- */
- assert(packed_consts.smear < 8);
}
} else {
/* Turn the byte offset into a dword offset. */
break;
case GLSL_TYPE_SAMPLER:
+ case GLSL_TYPE_IMAGE:
case GLSL_TYPE_ATOMIC_UINT:
break;
/* sample index */
emit(MOV(fs_reg(MRF, base_mrf + mlen, BRW_REGISTER_TYPE_UD), sample_index));
mlen += reg_width;
- inst = emit(SHADER_OPCODE_TXF_MS, dst);
+ inst = emit(SHADER_OPCODE_TXF_CMS, dst);
break;
case ir_lod:
inst = emit(SHADER_OPCODE_LOD, dst);
inst->header_present = header_present;
inst->regs_written = 4;
- if (mlen > 11) {
- fail("Message length >11 disallowed by hardware\n");
+ if (mlen > MAX_SAMPLER_MESSAGE_SIZE) {
+ fail("Message length >" STRINGIFY(MAX_SAMPLER_MESSAGE_SIZE)
+ " disallowed by hardware\n");
}
return inst;
fs_inst *
fs_visitor::emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
fs_reg shadow_c, fs_reg lod, fs_reg lod2,
- fs_reg sample_index, fs_reg mcs)
+ fs_reg sample_index, fs_reg mcs, int sampler)
{
int reg_width = dispatch_width / 8;
bool header_present = false;
fs_reg payload = fs_reg(this, glsl_type::float_type);
fs_reg next = payload;
- if (ir->op == ir_tg4 || (ir->offset && ir->op != ir_txf)) {
+ if (ir->op == ir_tg4 || (ir->offset && ir->op != ir_txf) || sampler >= 16) {
/* For general texture offsets (no txf workaround), we need a header to
- * put them in. Note that for 16-wide we're making space for two actual
+ * put them in. Note that for SIMD16 we're making space for two actual
* hardware registers here, so the emit will have to fix up for this.
*
* * ir4_tg4 needs to place its channel select in the header,
* for interaction with ARB_texture_swizzle
+ *
+ * The sampler index is only 4-bits, so for larger sampler numbers we
+ * need to offset the Sampler State Pointer in the header.
*/
header_present = true;
next.reg_offset++;
break;
}
case ir_txs:
- emit(MOV(next.retype(BRW_REGISTER_TYPE_UD), lod));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_UD), lod));
next.reg_offset++;
break;
case ir_query_levels:
- emit(MOV(next.retype(BRW_REGISTER_TYPE_UD), fs_reg(0u)));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_UD), fs_reg(0u)));
next.reg_offset++;
break;
case ir_txf:
/* Unfortunately, the parameters for LD are intermixed: u, lod, v, r. */
- emit(MOV(next.retype(BRW_REGISTER_TYPE_D), coordinate));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_D), coordinate));
coordinate.reg_offset++;
next.reg_offset++;
- emit(MOV(next.retype(BRW_REGISTER_TYPE_D), lod));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_D), lod));
next.reg_offset++;
for (int i = 1; i < ir->coordinate->type->vector_elements; i++) {
- emit(MOV(next.retype(BRW_REGISTER_TYPE_D), coordinate));
+ emit(MOV(retype(next, 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));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_UD), sample_index));
next.reg_offset++;
/* data from the multisample control surface */
- emit(MOV(next.retype(BRW_REGISTER_TYPE_UD), mcs));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_UD), mcs));
next.reg_offset++;
/* there is no offsetting for this message; just copy in the integer
* texture coordinates
*/
for (int i = 0; i < ir->coordinate->type->vector_elements; i++) {
- emit(MOV(next.retype(BRW_REGISTER_TYPE_D), coordinate));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_D), 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));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_D), offset_value));
offset_value.reg_offset++;
next.reg_offset++;
}
case ir_txl: inst = emit(SHADER_OPCODE_TXL, dst, payload); break;
case ir_txd: inst = emit(SHADER_OPCODE_TXD, dst, payload); break;
case ir_txf: inst = emit(SHADER_OPCODE_TXF, dst, payload); break;
- case ir_txf_ms: inst = emit(SHADER_OPCODE_TXF_MS, dst, payload); break;
+ case ir_txf_ms: inst = emit(SHADER_OPCODE_TXF_CMS, dst, payload); break;
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;
inst->regs_written = 4;
virtual_grf_sizes[payload.reg] = next.reg_offset;
- if (inst->mlen > 11) {
- fail("Message length >11 disallowed by hardware\n");
+ if (inst->mlen > MAX_SAMPLER_MESSAGE_SIZE) {
+ fail("Message length >" STRINGIFY(MAX_SAMPLER_MESSAGE_SIZE)
+ " disallowed by hardware\n");
}
return inst;
};
if (dispatch_width == 16) {
- fail("rectangle scale uniform setup not supported on 16-wide\n");
+ fail("rectangle scale uniform setup not supported on SIMD16\n");
return coordinate;
}
- scale_x = fs_reg(UNIFORM, c->prog_data.nr_params);
- scale_y = fs_reg(UNIFORM, c->prog_data.nr_params + 1);
+ scale_x = fs_reg(UNIFORM, stage_prog_data->nr_params);
+ scale_y = fs_reg(UNIFORM, stage_prog_data->nr_params + 1);
GLuint index = _mesa_add_state_reference(params,
(gl_state_index *)tokens);
- c->prog_data.param[c->prog_data.nr_params++] =
+ stage_prog_data->param[stage_prog_data->nr_params++] =
&prog->Parameters->ParameterValues[index][0].f;
- c->prog_data.param[c->prog_data.nr_params++] =
+ stage_prog_data->param[stage_prog_data->nr_params++] =
&prog->Parameters->ParameterValues[index][1].f;
}
/* parameters are: u, v, r, lod; missing parameters are treated as zero */
for (int i = 0; i < ir->coordinate->type->vector_elements; i++) {
- emit(MOV(next.retype(BRW_REGISTER_TYPE_D), coordinate));
+ emit(MOV(retype(next, BRW_REGISTER_TYPE_D), coordinate));
coordinate.reg_offset++;
next.reg_offset++;
}
fs_inst *inst = emit(SHADER_OPCODE_TXF_MCS, dest, payload);
+ virtual_grf_sizes[payload.reg] = next.reg_offset;
inst->base_mrf = -1;
inst->mlen = next.reg_offset * reg_width;
inst->header_present = false;
if (brw->gen >= 7) {
inst = emit_texture_gen7(ir, dst, coordinate, shadow_comparitor,
- lod, lod2, sample_index, mcs);
+ lod, lod2, sample_index, mcs, sampler);
} else if (brw->gen >= 5) {
inst = emit_texture_gen5(ir, dst, coordinate, shadow_comparitor,
lod, lod2, sample_index);
}
}
+ if (brw->gen == 6 && ir->op == ir_tg4) {
+ emit_gen6_gather_wa(c->key.tex.gen6_gather_wa[sampler], dst);
+ }
+
swizzle_result(ir, dst, sampler);
}
/**
+ * Apply workarounds for Gen6 gather with UINT/SINT
+ */
+void
+fs_visitor::emit_gen6_gather_wa(uint8_t wa, fs_reg dst)
+{
+ if (!wa)
+ return;
+
+ int width = (wa & WA_8BIT) ? 8 : 16;
+
+ for (int i = 0; i < 4; i++) {
+ fs_reg dst_f = retype(dst, BRW_REGISTER_TYPE_F);
+ /* Convert from UNORM to UINT */
+ emit(MUL(dst_f, dst_f, fs_reg((float)((1 << width) - 1))));
+ emit(MOV(dst, dst_f));
+
+ if (wa & WA_SIGN) {
+ /* Reinterpret the UINT value as a signed INT value by
+ * shifting the sign bit into place, then shifting back
+ * preserving sign.
+ */
+ emit(SHL(dst, dst, fs_reg(32 - width)));
+ emit(ASR(dst, dst, fs_reg(32 - width)));
+ }
+
+ dst.reg_offset++;
+ }
+}
+
+/**
* Set up the gather channel based on the swizzle, for gather4.
*/
uint32_t
fs_visitor::visit(ir_if *ir)
{
if (brw->gen < 6 && dispatch_width == 16) {
- fail("Can't support (non-uniform) control flow on 16-wide\n");
+ fail("Can't support (non-uniform) control flow on SIMD16\n");
}
/* Don't point the annotation at the if statement, because then it plus
fs_visitor::visit(ir_loop *ir)
{
if (brw->gen < 6 && dispatch_width == 16) {
- fail("Can't support (non-uniform) control flow on 16-wide\n");
+ fail("Can't support (non-uniform) control flow on SIMD16\n");
}
this->base_ir = NULL;
ir->actual_parameters.get_head());
ir_variable *location = deref->variable_referenced();
unsigned surf_index = (c->prog_data.base.binding_table.abo_start +
- location->atomic.buffer_index);
+ location->data.atomic.buffer_index);
/* Calculate the surface offset */
fs_reg offset(this, glsl_type::uint_type);
fs_reg tmp(this, glsl_type::uint_type);
emit(MUL(tmp, this->result, ATOMIC_COUNTER_SIZE));
- emit(ADD(offset, tmp, location->atomic.offset));
+ emit(ADD(offset, tmp, location->data.atomic.offset));
} else {
- offset = location->atomic.offset;
+ offset = location->data.atomic.offset;
}
/* Emit the appropriate machine instruction */
inst->saturate = c->key.clamp_fragment_color;
pop_force_uncompressed();
- color.sechalf = true;
inst = emit(MOV(fs_reg(MRF, first_color_mrf + index + 4, color.type),
- color));
+ half(color, 1)));
inst->force_sechalf = true;
inst->saturate = c->key.clamp_fragment_color;
- color.sechalf = false;
}
}
}
bool src0_alpha_to_render_target = false;
if (dispatch_width == 16 && do_dual_src) {
- fail("GL_ARB_blend_func_extended not yet supported in 16-wide.");
+ fail("GL_ARB_blend_func_extended not yet supported in SIMD16.");
do_dual_src = false;
}
* thread message and on all dual-source messages."
*/
if (brw->gen >= 6 &&
- !this->fp->UsesKill &&
+ (brw->is_haswell || brw->gen >= 8 || !this->fp->UsesKill) &&
!do_dual_src &&
c->key.nr_color_regions == 1) {
header_present = false;
if (c->source_depth_to_render_target) {
if (brw->gen == 6 && dispatch_width == 16) {
/* For outputting oDepth on gen6, SIMD8 writes have to be
- * used. This would require 8-wide moves of each half to
+ * used. This would require SIMD8 moves of each half to
* message regs, kind of like pre-gen5 SIMD16 FB writes.
* Just bail on doing so for now.
*/
inst->mlen = nr - base_mrf;
inst->eot = true;
inst->header_present = header_present;
+ if ((brw->gen >= 8 || brw->is_haswell) && fp->UsesKill) {
+ inst->predicate = BRW_PREDICATE_NORMAL;
+ inst->flag_subreg = 1;
+ }
c->prog_data.dual_src_blend = true;
this->current_annotation = NULL;
inst->mlen = nr - base_mrf;
inst->eot = eot;
inst->header_present = header_present;
+ if ((brw->gen >= 8 || brw->is_haswell) && fp->UsesKill) {
+ inst->predicate = BRW_PREDICATE_NORMAL;
+ inst->flag_subreg = 1;
+ }
}
if (c->key.nr_color_regions == 0) {
inst->mlen = nr - base_mrf;
inst->eot = true;
inst->header_present = header_present;
+ if ((brw->gen >= 8 || brw->is_haswell) && fp->UsesKill) {
+ inst->predicate = BRW_PREDICATE_NORMAL;
+ inst->flag_subreg = 1;
+ }
}
this->current_annotation = NULL;
this->virtual_grf_start = NULL;
this->virtual_grf_end = NULL;
this->live_intervals = NULL;
+ this->regs_live_at_ip = NULL;
this->params_remap = NULL;
this->nr_params_remap = 0;