this->shadow_compare = false;
this->ir = NULL;
this->urb_write_flags = BRW_URB_WRITE_NO_FLAGS;
- this->header_present = false;
+ this->header_size = 0;
this->flag_subreg = 0;
this->mlen = 0;
this->base_mrf = 0;
vec4_visitor::op(const dst_reg &dst, const src_reg &src0, \
const src_reg &src1, const src_reg &src2) \
{ \
- assert(brw->gen >= 6); \
+ assert(devinfo->gen >= 6); \
return new(mem_ctx) vec4_instruction(BRW_OPCODE_##op, dst, \
src0, src1, src2); \
}
vec4_visitor::IF(src_reg src0, src_reg src1,
enum brw_conditional_mod condition)
{
- assert(brw->gen == 6);
+ assert(devinfo->gen == 6);
vec4_instruction *inst;
src_reg
vec4_visitor::fix_math_operand(src_reg src)
{
- if (brw->gen < 6 || brw->gen >= 8 || src.file == BAD_FILE)
+ if (devinfo->gen < 6 || devinfo->gen >= 8 || src.file == BAD_FILE)
return src;
/* The gen6 math instruction ignores the source modifiers --
* can't use.
*/
- if (brw->gen == 7 && src.file != IMM)
+ if (devinfo->gen == 7 && src.file != IMM)
return src;
dst_reg expanded = dst_reg(this, glsl_type::vec4_type);
vec4_instruction *math =
emit(opcode, dst, fix_math_operand(src0), fix_math_operand(src1));
- if (brw->gen == 6 && dst.writemask != WRITEMASK_XYZW) {
+ if (devinfo->gen == 6 && dst.writemask != WRITEMASK_XYZW) {
/* MATH on Gen6 must be align1, so we can't do writemasks. */
math->dst = dst_reg(this, glsl_type::vec4_type);
math->dst.type = dst.type;
emit(MOV(dst, src_reg(math->dst)));
- } else if (brw->gen < 6) {
+ } else if (devinfo->gen < 6) {
math->base_mrf = 1;
math->mlen = src1.file == BAD_FILE ? 1 : 2;
}
void
vec4_visitor::emit_pack_half_2x16(dst_reg dst, src_reg src0)
{
- if (brw->gen < 7) {
+ if (devinfo->gen < 7) {
unreachable("ir_unop_pack_half_2x16 should be lowered");
}
void
vec4_visitor::emit_unpack_half_2x16(dst_reg dst, src_reg src0)
{
- if (brw->gen < 7) {
+ if (devinfo->gen < 7) {
unreachable("ir_unop_unpack_half_2x16 should be lowered");
}
}
}
-
-static int
-type_size(const struct glsl_type *type)
+/**
+ * Returns the minimum number of vec4 elements needed to pack a type.
+ *
+ * For simple types, it will return 1 (a single vec4); for matrices, the
+ * number of columns; for array and struct, the sum of the vec4_size of
+ * each of its elements; and for sampler and atomic, zero.
+ *
+ * This method is useful to calculate how much register space is needed to
+ * store a particular type.
+ */
+int
+vec4_visitor::type_size(const struct glsl_type *type)
{
unsigned int i;
int size;
size += type_size(type->fields.structure[i].type);
}
return size;
+ case GLSL_TYPE_SUBROUTINE:
+ return 1;
+
case GLSL_TYPE_SAMPLER:
/* Samplers take up no register space, since they're baked in at
* link time.
init();
this->file = GRF;
- this->reg = v->alloc.allocate(type_size(type));
+ this->reg = v->alloc.allocate(v->type_size(type));
if (type->is_array() || type->is_record()) {
this->swizzle = BRW_SWIZZLE_NOOP;
init();
this->file = GRF;
- this->reg = v->alloc.allocate(type_size(type) * size);
+ this->reg = v->alloc.allocate(v->type_size(type) * size);
this->swizzle = BRW_SWIZZLE_NOOP;
init();
this->file = GRF;
- this->reg = v->alloc.allocate(type_size(type));
+ this->reg = v->alloc.allocate(v->type_size(type));
if (type->is_array() || type->is_record()) {
this->writemask = WRITEMASK_XYZW;
this->type = brw_type_for_base_type(type);
}
+void
+vec4_visitor::setup_vector_uniform_values(const gl_constant_value *values,
+ unsigned n)
+{
+ static const gl_constant_value zero = { 0 };
+
+ for (unsigned i = 0; i < n; ++i)
+ stage_prog_data->param[4 * uniforms + i] = &values[i];
+
+ for (unsigned i = n; i < 4; ++i)
+ stage_prog_data->param[4 * uniforms + i] = &zero;
+
+ uniform_vector_size[uniforms++] = n;
+}
+
/* Our support for uniforms is piggy-backed on the struct
* gl_fragment_program, because that's where the values actually
* get stored, rather than in some global gl_shader_program uniform
* order we'd walk the type, so walk the list of storage and find anything
* with our name, or the prefix of a component that starts with our name.
*/
- for (unsigned u = 0; u < shader_prog->NumUserUniformStorage; u++) {
+ for (unsigned u = 0; u < shader_prog->NumUniformStorage; u++) {
struct gl_uniform_storage *storage = &shader_prog->UniformStorage[u];
+ if (storage->builtin)
+ continue;
+
if (strncmp(ir->name, storage->name, namelen) != 0 ||
(storage->name[namelen] != 0 &&
storage->name[namelen] != '.' &&
continue;
}
- gl_constant_value *components = storage->storage;
- unsigned vector_count = (MAX2(storage->array_elements, 1) *
- storage->type->matrix_columns);
-
- for (unsigned s = 0; s < vector_count; s++) {
- assert(uniforms < uniform_array_size);
- uniform_vector_size[uniforms] = storage->type->vector_elements;
+ const unsigned vector_count = (MAX2(storage->array_elements, 1) *
+ storage->type->matrix_columns);
+ const unsigned vector_size = storage->type->vector_elements;
- int i;
- for (i = 0; i < uniform_vector_size[uniforms]; i++) {
- stage_prog_data->param[uniforms * 4 + i] = components;
- components++;
- }
- for (; i < 4; i++) {
- static gl_constant_value zero = { 0.0 };
- stage_prog_data->param[uniforms * 4 + i] = &zero;
- }
-
- uniforms++;
- }
+ for (unsigned s = 0; s < vector_count; s++)
+ setup_vector_uniform_values(&storage->storage[s * vector_size],
+ vector_size);
}
}
void
-vec4_visitor::setup_uniform_clipplane_values()
+vec4_visitor::setup_uniform_clipplane_values(gl_clip_plane *clip_planes)
{
- gl_clip_plane *clip_planes = brw_select_clip_planes(ctx);
-
for (int i = 0; i < key->nr_userclip_plane_consts; ++i) {
assert(this->uniforms < uniform_array_size);
this->uniform_vector_size[this->uniforms] = 4;
break;
case ir_binop_logic_xor:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
src_reg temp = src_reg(this, ir->type);
emit(XOR(dst_reg(temp), op[0], op[1]));
inst = emit(AND(dst_null_d(), temp, src_reg(1)));
break;
case ir_binop_logic_or:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
src_reg temp = src_reg(this, ir->type);
emit(OR(dst_reg(temp), op[0], op[1]));
inst = emit(AND(dst_null_d(), temp, src_reg(1)));
break;
case ir_binop_logic_and:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
src_reg temp = src_reg(this, ir->type);
emit(AND(dst_reg(temp), op[0], op[1]));
inst = emit(AND(dst_null_d(), temp, src_reg(1)));
break;
case ir_unop_f2b:
- if (brw->gen >= 6) {
+ if (devinfo->gen >= 6) {
emit(CMP(dst_null_d(), op[0], src_reg(0.0f), BRW_CONDITIONAL_NZ));
} else {
inst = emit(MOV(dst_null_f(), op[0]));
break;
case ir_unop_i2b:
- if (brw->gen >= 6) {
+ if (devinfo->gen >= 6) {
emit(CMP(dst_null_d(), op[0], src_reg(0), BRW_CONDITIONAL_NZ));
} else {
inst = emit(MOV(dst_null_d(), op[0]));
break;
case ir_binop_all_equal:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(expr->operands[0], &op[0]);
resolve_bool_comparison(expr->operands[1], &op[1]);
}
break;
case ir_binop_any_nequal:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(expr->operands[0], &op[0]);
resolve_bool_comparison(expr->operands[1], &op[1]);
}
break;
case ir_unop_any:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(expr->operands[0], &op[0]);
}
inst = emit(CMP(dst_null_d(), op[0], src_reg(0), BRW_CONDITIONAL_NZ));
case ir_binop_lequal:
case ir_binop_equal:
case ir_binop_nequal:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(expr->operands[0], &op[0]);
resolve_bool_comparison(expr->operands[1], &op[1]);
}
for (int i = 0; i < type_size(ir->type); i++) {
output_reg[ir->data.location + i] = *reg;
output_reg[ir->data.location + i].reg_offset = i;
- output_reg[ir->data.location + i].type =
- brw_type_for_base_type(ir->type->get_scalar_type());
output_reg_annotation[ir->data.location + i] = ir->name;
}
break;
* Some uniforms, such as samplers and atomic counters, have no actual
* storage, so we should ignore them.
*/
- if (ir->is_in_uniform_block() || type_size(ir->type) == 0)
+ if (ir->is_in_buffer_block() || type_size(ir->type) == 0)
return;
/* Track how big the whole uniform variable is, in case we need to put a
break;
case ir_var_system_value:
- reg = make_reg_for_system_value(ir);
+ reg = make_reg_for_system_value(ir->data.location, ir->type);
break;
default:
vec4_visitor::try_emit_mad(ir_expression *ir)
{
/* 3-src instructions were introduced in gen6. */
- if (brw->gen < 6)
+ if (devinfo->gen < 6)
return false;
/* MAD can only handle floating-point data. */
* false. Early hardware only sets the least significant bit, and
* leaves the other bits undefined. So we can't use it.
*/
- if (brw->gen < 6)
+ if (devinfo->gen < 6)
return false;
ir_expression *const cmp = ir->operands[0]->as_expression();
{
vec4_instruction *inst;
- if (brw->gen >= 6) {
+ if (devinfo->gen >= 6) {
inst = emit(BRW_OPCODE_SEL, dst, src0, src1);
inst->conditional_mod = conditionalmod;
} else {
vec4_visitor::emit_lrp(const dst_reg &dst,
const src_reg &x, const src_reg &y, const src_reg &a)
{
- if (brw->gen >= 6) {
+ if (devinfo->gen >= 6) {
/* Note that the instruction's argument order is reversed from GLSL
* and the IR.
*/
}
}
+/**
+ * Emits the instructions needed to perform a pull constant load. before_block
+ * and before_inst can be NULL in which case the instruction will be appended
+ * to the end of the instruction list.
+ */
+void
+vec4_visitor::emit_pull_constant_load_reg(dst_reg dst,
+ src_reg surf_index,
+ src_reg offset_reg,
+ bblock_t *before_block,
+ vec4_instruction *before_inst)
+{
+ assert((before_inst == NULL && before_block == NULL) ||
+ (before_inst && before_block));
+
+ vec4_instruction *pull;
+
+ if (devinfo->gen >= 9) {
+ /* Gen9+ needs a message header in order to use SIMD4x2 mode */
+ src_reg header(this, glsl_type::uvec4_type, 2);
+
+ pull = new(mem_ctx)
+ vec4_instruction(VS_OPCODE_SET_SIMD4X2_HEADER_GEN9,
+ dst_reg(header));
+
+ if (before_inst)
+ emit_before(before_block, before_inst, pull);
+ else
+ emit(pull);
+
+ dst_reg index_reg = retype(offset(dst_reg(header), 1),
+ offset_reg.type);
+ pull = MOV(writemask(index_reg, WRITEMASK_X), offset_reg);
+
+ if (before_inst)
+ emit_before(before_block, before_inst, pull);
+ else
+ emit(pull);
+
+ pull = new(mem_ctx) vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD_GEN7,
+ dst,
+ surf_index,
+ header);
+ pull->mlen = 2;
+ pull->header_size = 1;
+ } else if (devinfo->gen >= 7) {
+ dst_reg grf_offset = dst_reg(this, glsl_type::int_type);
+
+ grf_offset.type = offset_reg.type;
+
+ pull = MOV(grf_offset, offset_reg);
+
+ if (before_inst)
+ emit_before(before_block, before_inst, pull);
+ else
+ emit(pull);
+
+ pull = new(mem_ctx) vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD_GEN7,
+ dst,
+ surf_index,
+ src_reg(grf_offset));
+ pull->mlen = 1;
+ } else {
+ pull = new(mem_ctx) vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD,
+ dst,
+ surf_index,
+ offset_reg);
+ pull->base_mrf = 14;
+ pull->mlen = 1;
+ }
+
+ if (before_inst)
+ emit_before(before_block, before_inst, pull);
+ else
+ emit(pull);
+}
+
+src_reg
+vec4_visitor::emit_uniformize(const src_reg &src)
+{
+ const src_reg chan_index(this, glsl_type::uint_type);
+ const dst_reg dst = retype(dst_reg(this, glsl_type::uint_type),
+ src.type);
+
+ emit(SHADER_OPCODE_FIND_LIVE_CHANNEL, dst_reg(chan_index))
+ ->force_writemask_all = true;
+ emit(SHADER_OPCODE_BROADCAST, dst, src, chan_index)
+ ->force_writemask_all = true;
+
+ return src_reg(dst);
+}
+
void
vec4_visitor::visit(ir_expression *ir)
{
case ir_unop_log:
unreachable("not reached: should be handled by ir_explog_to_explog2");
case ir_unop_sin:
- case ir_unop_sin_reduced:
emit_math(SHADER_OPCODE_SIN, result_dst, op[0]);
break;
case ir_unop_cos:
- case ir_unop_cos_reduced:
emit_math(SHADER_OPCODE_COS, result_dst, op[0]);
break;
case ir_unop_noise:
unreachable("not reached: should be handled by lower_noise");
+ case ir_unop_subroutine_to_int:
+ emit(MOV(result_dst, op[0]));
+ break;
+
case ir_binop_add:
emit(ADD(result_dst, op[0], op[1]));
break;
unreachable("not reached: should be handled by ir_sub_to_add_neg");
case ir_binop_mul:
- if (brw->gen < 8 && ir->type->is_integer()) {
+ if (devinfo->gen < 8 && ir->type->is_integer()) {
/* For integer multiplication, the MUL uses the low 16 bits of one of
* the operands (src0 through SNB, src1 on IVB and later). The MACH
* accumulates in the contribution of the upper 16 bits of that
* 16 bits, though, we can just emit a single MUL.
*/
if (ir->operands[0]->is_uint16_constant()) {
- if (brw->gen < 7)
+ if (devinfo->gen < 7)
emit(MUL(result_dst, op[0], op[1]));
else
emit(MUL(result_dst, op[1], op[0]));
} else if (ir->operands[1]->is_uint16_constant()) {
- if (brw->gen < 7)
+ if (devinfo->gen < 7)
emit(MUL(result_dst, op[1], op[0]));
else
emit(MUL(result_dst, op[0], op[1]));
assert(ir->type->is_integer());
emit_math(SHADER_OPCODE_INT_QUOTIENT, result_dst, op[0], op[1]);
break;
- case ir_binop_carry: {
- struct brw_reg acc = retype(brw_acc_reg(8), BRW_REGISTER_TYPE_UD);
- emit(ADDC(dst_null_ud(), op[0], op[1]));
- emit(MOV(result_dst, src_reg(acc)));
- break;
- }
- case ir_binop_borrow: {
- struct brw_reg acc = retype(brw_acc_reg(8), BRW_REGISTER_TYPE_UD);
+ case ir_binop_carry:
+ unreachable("Should have been lowered by carry_to_arith().");
+
+ case ir_binop_borrow:
+ unreachable("Should have been lowered by borrow_to_arith().");
- emit(SUBB(dst_null_ud(), op[0], op[1]));
- emit(MOV(result_dst, src_reg(acc)));
- break;
- }
case ir_binop_mod:
/* Floating point should be lowered by MOD_TO_FLOOR in the compiler. */
assert(ir->type->is_integer());
case ir_binop_gequal:
case ir_binop_equal:
case ir_binop_nequal: {
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(ir->operands[0], &op[0]);
resolve_bool_comparison(ir->operands[1], &op[1]);
}
}
case ir_binop_all_equal:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(ir->operands[0], &op[0]);
resolve_bool_comparison(ir->operands[1], &op[1]);
}
ir->operands[1]->type->is_vector()) {
emit(CMP(dst_null_d(), op[0], op[1], BRW_CONDITIONAL_Z));
emit(MOV(result_dst, src_reg(0)));
- inst = emit(MOV(result_dst, src_reg((int)ctx->Const.UniformBooleanTrue)));
+ inst = emit(MOV(result_dst, src_reg(~0)));
inst->predicate = BRW_PREDICATE_ALIGN16_ALL4H;
} else {
emit(CMP(result_dst, op[0], op[1], BRW_CONDITIONAL_Z));
}
break;
case ir_binop_any_nequal:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(ir->operands[0], &op[0]);
resolve_bool_comparison(ir->operands[1], &op[1]);
}
emit(CMP(dst_null_d(), op[0], op[1], BRW_CONDITIONAL_NZ));
emit(MOV(result_dst, src_reg(0)));
- inst = emit(MOV(result_dst, src_reg((int)ctx->Const.UniformBooleanTrue)));
+ inst = emit(MOV(result_dst, src_reg(~0)));
inst->predicate = BRW_PREDICATE_ALIGN16_ANY4H;
} else {
emit(CMP(result_dst, op[0], op[1], BRW_CONDITIONAL_NZ));
break;
case ir_unop_any:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(ir->operands[0], &op[0]);
}
emit(CMP(dst_null_d(), op[0], src_reg(0), BRW_CONDITIONAL_NZ));
emit(MOV(result_dst, src_reg(0)));
- inst = emit(MOV(result_dst, src_reg((int)ctx->Const.UniformBooleanTrue)));
+ inst = emit(MOV(result_dst, src_reg(~0)));
inst->predicate = BRW_PREDICATE_ALIGN16_ANY4H;
break;
emit(MOV(result_dst, op[0]));
break;
case ir_unop_b2i:
- emit(AND(result_dst, op[0], src_reg(1)));
- break;
case ir_unop_b2f:
- if (brw->gen <= 5) {
+ if (devinfo->gen <= 5) {
resolve_bool_comparison(ir->operands[0], &op[0]);
}
- op[0].type = BRW_REGISTER_TYPE_D;
- result_dst.type = BRW_REGISTER_TYPE_D;
- emit(AND(result_dst, op[0], src_reg(0x3f800000u)));
- result_dst.type = BRW_REGISTER_TYPE_F;
+ emit(MOV(result_dst, negate(op[0])));
break;
case ir_unop_f2b:
emit(CMP(result_dst, op[0], src_reg(0.0f), BRW_CONDITIONAL_NZ));
const_uniform_block->value.u[0]);
} else {
/* The block index is not a constant. Evaluate the index expression
- * per-channel and add the base UBO index; the generator will select
- * a value from any live channel.
+ * per-channel and add the base UBO index; we have to select a value
+ * from any live channel.
*/
surf_index = src_reg(this, glsl_type::uint_type);
emit(ADD(dst_reg(surf_index), op[0],
src_reg(prog_data->base.binding_table.ubo_start)));
+ surf_index = emit_uniformize(surf_index);
/* Assume this may touch any UBO. It would be nice to provide
* a tighter bound, but the array information is already lowered away.
}
if (const_offset_ir) {
- if (brw->gen >= 8) {
+ if (devinfo->gen >= 8) {
/* Store the offset in a GRF so we can send-from-GRF. */
offset = src_reg(this, glsl_type::int_type);
emit(MOV(dst_reg(offset), src_reg(const_offset / 16)));
}
} else {
offset = src_reg(this, glsl_type::uint_type);
- emit(SHR(dst_reg(offset), op[1], src_reg(4)));
+ emit(SHR(dst_reg(offset), op[1], src_reg(4u)));
}
- if (brw->gen >= 7) {
- dst_reg grf_offset = dst_reg(this, glsl_type::int_type);
-
- /* We have to use a message header on Skylake to get SIMD4x2 mode.
- * Reserve space for the register.
- */
- if (brw->gen >= 9) {
- grf_offset.reg_offset++;
- alloc.sizes[grf_offset.reg] = 2;
- }
-
- grf_offset.type = offset.type;
-
- emit(MOV(grf_offset, offset));
-
- vec4_instruction *pull =
- emit(new(mem_ctx) vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD_GEN7,
- dst_reg(packed_consts),
- surf_index,
- src_reg(grf_offset)));
- pull->mlen = 1;
- } else {
- vec4_instruction *pull =
- emit(new(mem_ctx) vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD,
- dst_reg(packed_consts),
- surf_index,
- offset));
- pull->base_mrf = 14;
- pull->mlen = 1;
- }
+ emit_pull_constant_load_reg(dst_reg(packed_consts),
+ surf_index,
+ offset,
+ NULL, NULL /* before_block/inst */);
packed_consts.swizzle = brw_swizzle_for_size(ir->type->vector_elements);
packed_consts.swizzle += BRW_SWIZZLE4(const_offset % 16 / 4,
const_offset % 16 / 4,
const_offset % 16 / 4);
- /* UBO bools are any nonzero int. We need to convert them to use the
- * value of true stored in ctx->Const.UniformBooleanTrue.
- */
+ /* UBO bools are any nonzero int. We need to convert them to 0/~0. */
if (ir->type->base_type == GLSL_TYPE_BOOL) {
emit(CMP(result_dst, packed_consts, src_reg(0u),
BRW_CONDITIONAL_NZ));
emit(MOV(*dst, src_reg(ir->value.u[i])));
break;
case GLSL_TYPE_BOOL:
- emit(MOV(*dst,
- src_reg(ir->value.b[i] != 0 ? (int)ctx->Const.UniformBooleanTrue
- : 0)));
+ emit(MOV(*dst, src_reg(ir->value.b[i] != 0 ? ~0 : 0)));
break;
default:
unreachable("Non-float/uint/int/bool constant");
emit_untyped_atomic(BRW_AOP_PREDEC, surf_index, dst, offset,
src_reg(), src_reg());
}
+
+ brw_mark_surface_used(stage_prog_data, surf_index);
}
void
new(mem_ctx) vec4_instruction(SHADER_OPCODE_TXF_MCS,
dst_reg(this, glsl_type::uvec4_type));
inst->base_mrf = 2;
- inst->mlen = 1;
inst->src[1] = sampler;
+ int param_base;
+
+ if (devinfo->gen >= 9) {
+ /* Gen9+ needs a message header in order to use SIMD4x2 mode */
+ vec4_instruction *header_inst = new(mem_ctx)
+ vec4_instruction(VS_OPCODE_SET_SIMD4X2_HEADER_GEN9,
+ dst_reg(MRF, inst->base_mrf));
+
+ emit(header_inst);
+
+ inst->mlen = 2;
+ inst->header_size = 1;
+ param_base = inst->base_mrf + 1;
+ } else {
+ inst->mlen = 1;
+ param_base = inst->base_mrf;
+ }
+
/* parameters are: u, v, r, lod; lod will always be zero due to api restrictions */
- int param_base = inst->base_mrf;
int coord_mask = (1 << ir->coordinate->type->vector_elements) - 1;
int zero_mask = 0xf & ~coord_mask;
}
static bool
-is_high_sampler(struct brw_context *brw, src_reg sampler)
+is_high_sampler(const struct brw_device_info *devinfo, src_reg sampler)
{
- if (brw->gen < 8 && !brw->is_haswell)
+ if (devinfo->gen < 8 && !devinfo->is_haswell)
return false;
return sampler.file != IMM || sampler.fixed_hw_reg.dw1.ud >= 16;
->array->type->array_size();
uint32_t max_used = sampler + array_size - 1;
- if (ir->op == ir_tg4 && brw->gen < 8) {
+ if (ir->op == ir_tg4 && devinfo->gen < 8) {
max_used += prog_data->base.binding_table.gather_texture_start;
} else {
max_used += prog_data->base.binding_table.texture_start;
/* Emit code to evaluate the actual indexing expression */
nonconst_sampler_index->accept(this);
- dst_reg temp(this, glsl_type::uint_type);
- emit(ADD(temp, this->result, src_reg(sampler)))
- ->force_writemask_all = true;
- sampler_reg = src_reg(temp);
+ src_reg temp(this, glsl_type::uint_type);
+ emit(ADD(dst_reg(temp), this->result, src_reg(sampler)));
+ sampler_reg = emit_uniformize(temp);
} else {
/* Single sampler, or constant array index; the indexing expression
* is just an immediate.
sample_index = this->result;
sample_index_type = ir->lod_info.sample_index->type;
- if (brw->gen >= 7 && key->tex.compressed_multisample_layout_mask & (1<<sampler))
+ if (devinfo->gen >= 7 && key->tex.compressed_multisample_layout_mask & (1<<sampler))
mcs = emit_mcs_fetch(ir, coordinate, sampler_reg);
else
mcs = src_reg(0u);
if (ir->offset != NULL && !has_nonconstant_offset) {
inst->offset =
- brw_texture_offset(ctx, ir->offset->as_constant()->value.i,
+ brw_texture_offset(ir->offset->as_constant()->value.i,
ir->offset->type->vector_elements);
}
* - Gather channel selection
* - Sampler indices too large to fit in a 4-bit value.
*/
- inst->header_present =
- brw->gen < 5 || brw->gen >= 9 ||
- inst->offset != 0 || ir->op == ir_tg4 ||
- is_high_sampler(brw, sampler_reg);
+ inst->header_size =
+ (devinfo->gen < 5 || devinfo->gen >= 9 ||
+ inst->offset != 0 || ir->op == ir_tg4 ||
+ is_high_sampler(devinfo, sampler_reg)) ? 1 : 0;
inst->base_mrf = 2;
- inst->mlen = inst->header_present + 1; /* always at least one */
+ inst->mlen = inst->header_size + 1; /* always at least one */
inst->dst.writemask = WRITEMASK_XYZW;
inst->shadow_compare = ir->shadow_comparitor != NULL;
inst->src[1] = sampler_reg;
/* MRF for the first parameter */
- int param_base = inst->base_mrf + inst->header_present;
+ int param_base = inst->base_mrf + inst->header_size;
if (ir->op == ir_txs || ir->op == ir_query_levels) {
- int writemask = brw->gen == 4 ? WRITEMASK_W : WRITEMASK_X;
+ int writemask = devinfo->gen == 4 ? WRITEMASK_W : WRITEMASK_X;
emit(MOV(dst_reg(MRF, param_base, lod_type, writemask), lod));
} else {
/* Load the coordinate */
/* Load the LOD info */
if (ir->op == ir_tex || ir->op == ir_txl) {
int mrf, writemask;
- if (brw->gen >= 5) {
+ if (devinfo->gen >= 5) {
mrf = param_base + 1;
if (ir->shadow_comparitor) {
writemask = WRITEMASK_Y;
writemask = WRITEMASK_X;
inst->mlen++;
}
- } else /* brw->gen == 4 */ {
+ } else /* devinfo->gen == 4 */ {
mrf = param_base;
writemask = WRITEMASK_W;
}
} else if (ir->op == ir_txf_ms) {
emit(MOV(dst_reg(MRF, param_base + 1, sample_index_type, WRITEMASK_X),
sample_index));
- if (brw->gen >= 7) {
+ if (devinfo->gen >= 7) {
/* MCS data is in the first channel of `mcs`, but we need to get it into
* the .y channel of the second vec4 of params, so replicate .x across
* the whole vec4 and then mask off everything except .y
} else if (ir->op == ir_txd) {
const glsl_type *type = lod_type;
- if (brw->gen >= 5) {
+ if (devinfo->gen >= 5) {
dPdx.swizzle = BRW_SWIZZLE4(SWIZZLE_X,SWIZZLE_X,SWIZZLE_Y,SWIZZLE_Y);
dPdy.swizzle = BRW_SWIZZLE4(SWIZZLE_X,SWIZZLE_X,SWIZZLE_Y,SWIZZLE_Y);
emit(MOV(dst_reg(MRF, param_base + 1, type, WRITEMASK_XZ), dPdx));
shadow_comparitor));
}
}
- } else /* brw->gen == 4 */ {
+ } else /* devinfo->gen == 4 */ {
emit(MOV(dst_reg(MRF, param_base + 1, type, WRITEMASK_XYZ), dPdx));
emit(MOV(dst_reg(MRF, param_base + 2, type, WRITEMASK_XYZ), dPdy));
inst->mlen += 2;
}
}
- if (brw->gen == 6 && ir->op == ir_tg4) {
+ if (devinfo->gen == 6 && ir->op == ir_tg4) {
emit_gen6_gather_wa(key->tex.gen6_gather_wa[sampler], inst->dst);
}
*/
this->base_ir = ir->condition;
- if (brw->gen == 6) {
+ if (devinfo->gen == 6) {
emit_if_gen6(ir);
} else {
enum brw_predicate predicate;
unreachable("not reached");
}
+void
+vec4_visitor::visit(ir_barrier *)
+{
+ unreachable("not reached");
+}
+
void
vec4_visitor::emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
dst_reg dst, src_reg offset,
* unused channels will be masked out.
*/
vec4_instruction *inst = emit(SHADER_OPCODE_UNTYPED_ATOMIC, dst,
- src_reg(atomic_op), src_reg(surf_index));
- inst->base_mrf = 0;
+ brw_message_reg(0),
+ src_reg(surf_index), src_reg(atomic_op));
inst->mlen = mlen;
}
* untyped surface read message, but that's OK because unused
* channels will be masked out.
*/
- vec4_instruction *inst = emit(SHADER_OPCODE_UNTYPED_SURFACE_READ,
- dst, src_reg(surf_index));
- inst->base_mrf = 0;
+ vec4_instruction *inst = emit(SHADER_OPCODE_UNTYPED_SURFACE_READ, dst,
+ brw_message_reg(0),
+ src_reg(surf_index), src_reg(1));
inst->mlen = 1;
}
void
vec4_visitor::emit_psiz_and_flags(dst_reg reg)
{
- if (brw->gen < 6 &&
+ if (devinfo->gen < 6 &&
((prog_data->vue_map.slots_valid & VARYING_BIT_PSIZ) ||
- key->userclip_active || brw->has_negative_rhw_bug)) {
+ key->userclip_active || devinfo->has_negative_rhw_bug)) {
dst_reg header1 = dst_reg(this, glsl_type::uvec4_type);
dst_reg header1_w = header1;
header1_w.writemask = WRITEMASK_W;
* Later, clipping will detect ucp[6] and ensure the primitive is
* clipped against all fixed planes.
*/
- if (brw->has_negative_rhw_bug) {
+ if (devinfo->has_negative_rhw_bug) {
src_reg ndc_w = src_reg(output_reg[BRW_VARYING_SLOT_NDC]);
ndc_w.swizzle = BRW_SWIZZLE_WWWW;
emit(CMP(dst_null_f(), ndc_w, src_reg(0.0f), BRW_CONDITIONAL_L));
vec4_instruction *inst;
inst = emit(OR(header1_w, src_reg(header1_w), src_reg(1u << 6)));
inst->predicate = BRW_PREDICATE_NORMAL;
+ output_reg[BRW_VARYING_SLOT_NDC].type = BRW_REGISTER_TYPE_F;
inst = emit(MOV(output_reg[BRW_VARYING_SLOT_NDC], src_reg(0.0f)));
inst->predicate = BRW_PREDICATE_NORMAL;
}
emit(MOV(retype(reg, BRW_REGISTER_TYPE_UD), src_reg(header1)));
- } else if (brw->gen < 6) {
+ } else if (devinfo->gen < 6) {
emit(MOV(retype(reg, BRW_REGISTER_TYPE_UD), 0u));
} else {
emit(MOV(retype(reg, BRW_REGISTER_TYPE_D), src_reg(0)));
if (prog_data->vue_map.slots_valid & VARYING_BIT_PSIZ) {
dst_reg reg_w = reg;
reg_w.writemask = WRITEMASK_W;
- emit(MOV(reg_w, src_reg(output_reg[VARYING_SLOT_PSIZ])));
+ src_reg reg_as_src = src_reg(output_reg[VARYING_SLOT_PSIZ]);
+ reg_as_src.type = reg_w.type;
+ reg_as_src.swizzle = brw_swizzle_for_size(1);
+ emit(MOV(reg_w, reg_as_src));
}
if (prog_data->vue_map.slots_valid & VARYING_BIT_LAYER) {
dst_reg reg_y = reg;
reg_y.writemask = WRITEMASK_Y;
reg_y.type = BRW_REGISTER_TYPE_D;
+ output_reg[VARYING_SLOT_LAYER].type = reg_y.type;
emit(MOV(reg_y, src_reg(output_reg[VARYING_SLOT_LAYER])));
}
if (prog_data->vue_map.slots_valid & VARYING_BIT_VIEWPORT) {
dst_reg reg_z = reg;
reg_z.writemask = WRITEMASK_Z;
reg_z.type = BRW_REGISTER_TYPE_D;
+ output_reg[VARYING_SLOT_VIEWPORT].type = reg_z.type;
emit(MOV(reg_z, src_reg(output_reg[VARYING_SLOT_VIEWPORT])));
}
}
vec4_instruction *
vec4_visitor::emit_generic_urb_slot(dst_reg reg, int varying)
{
- assert (varying < VARYING_SLOT_MAX);
- reg.type = output_reg[varying].type;
+ assert(varying < VARYING_SLOT_MAX);
+ assert(output_reg[varying].type == reg.type);
current_annotation = output_reg_annotation[varying];
/* Copy the register, saturating if necessary */
return emit(MOV(reg, src_reg(output_reg[varying])));
vec4_visitor::emit_urb_slot(dst_reg reg, int varying)
{
reg.type = BRW_REGISTER_TYPE_F;
+ output_reg[varying].type = reg.type;
switch (varying) {
case VARYING_SLOT_PSIZ:
}
static int
-align_interleaved_urb_mlen(struct brw_context *brw, int mlen)
+align_interleaved_urb_mlen(const struct brw_device_info *devinfo, int mlen)
{
- if (brw->gen >= 6) {
+ if (devinfo->gen >= 6) {
/* URB data written (does not include the message header reg) must
* be a multiple of 256 bits, or 2 VS registers. See vol5c.5,
* section 5.4.3.2.2: URB_INTERLEAVED.
*/
emit_urb_write_header(mrf++);
- if (brw->gen < 6) {
+ if (devinfo->gen < 6) {
emit_ndc_computation();
}
current_annotation = "URB write";
vec4_instruction *inst = emit_urb_write_opcode(complete);
inst->base_mrf = base_mrf;
- inst->mlen = align_interleaved_urb_mlen(brw, mrf - base_mrf);
+ inst->mlen = align_interleaved_urb_mlen(devinfo, mrf - base_mrf);
inst->offset += offset;
} while(!complete);
}
/* Pre-gen6, the message header uses byte offsets instead of vec4
* (16-byte) offset units.
*/
- if (brw->gen < 6)
+ if (devinfo->gen < 6)
message_header_scale *= 16;
if (reladdr) {
/* Pre-gen6, the message header uses byte offsets instead of vec4
* (16-byte) offset units.
*/
- if (brw->gen < 6) {
+ if (devinfo->gen < 6) {
emit_before(block, inst, MUL(dst_reg(index), index, src_reg(16)));
}
return index;
- } else if (brw->gen >= 8) {
+ } else if (devinfo->gen >= 8) {
/* Store the offset in a GRF so we can send-from-GRF. */
src_reg offset = src_reg(this, glsl_type::int_type);
emit_before(block, inst, MOV(dst_reg(offset), src_reg(reg_offset)));
return offset;
} else {
- int message_header_scale = brw->gen < 6 ? 16 : 1;
+ int message_header_scale = devinfo->gen < 6 ? 16 : 1;
return src_reg(reg_offset * message_header_scale);
}
}
foreach_block_and_inst(block, vec4_instruction, inst, cfg) {
if (inst->dst.file == GRF && inst->dst.reladdr) {
if (scratch_loc[inst->dst.reg] == -1) {
- scratch_loc[inst->dst.reg] = c->last_scratch;
- c->last_scratch += this->alloc.sizes[inst->dst.reg];
+ scratch_loc[inst->dst.reg] = last_scratch;
+ last_scratch += this->alloc.sizes[inst->dst.reg];
}
for (src_reg *iter = inst->dst.reladdr;
iter->reladdr;
iter = iter->reladdr) {
if (iter->file == GRF && scratch_loc[iter->reg] == -1) {
- scratch_loc[iter->reg] = c->last_scratch;
- c->last_scratch += this->alloc.sizes[iter->reg];
+ scratch_loc[iter->reg] = last_scratch;
+ last_scratch += this->alloc.sizes[iter->reg];
}
}
}
iter->reladdr;
iter = iter->reladdr) {
if (iter->file == GRF && scratch_loc[iter->reg] == -1) {
- scratch_loc[iter->reg] = c->last_scratch;
- c->last_scratch += this->alloc.sizes[iter->reg];
+ scratch_loc[iter->reg] = last_scratch;
+ last_scratch += this->alloc.sizes[iter->reg];
}
}
}
src_reg index = src_reg(prog_data->base.binding_table.pull_constants_start);
src_reg offset = get_pull_constant_offset(block, inst, orig_src.reladdr,
reg_offset);
- vec4_instruction *load;
- if (brw->gen >= 7) {
- dst_reg grf_offset = dst_reg(this, glsl_type::int_type);
-
- /* We have to use a message header on Skylake to get SIMD4x2 mode.
- * Reserve space for the register.
- */
- if (brw->gen >= 9) {
- grf_offset.reg_offset++;
- alloc.sizes[grf_offset.reg] = 2;
- }
-
- grf_offset.type = offset.type;
- emit_before(block, inst, MOV(grf_offset, offset));
-
- load = new(mem_ctx) vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD_GEN7,
- temp, index, src_reg(grf_offset));
- load->mlen = 1;
- } else {
- load = new(mem_ctx) vec4_instruction(VS_OPCODE_PULL_CONSTANT_LOAD,
- temp, index, offset);
- load->base_mrf = 14;
- load->mlen = 1;
- }
- emit_before(block, inst, load);
+ emit_pull_constant_load_reg(temp,
+ index,
+ offset,
+ block, inst);
}
/**
void
vec4_visitor::resolve_bool_comparison(ir_rvalue *rvalue, src_reg *reg)
{
- assert(brw->gen <= 5);
+ assert(devinfo->gen <= 5);
if (!rvalue->type->is_boolean())
return;
*reg = neg_result;
}
-vec4_visitor::vec4_visitor(struct brw_context *brw,
- struct brw_vec4_compile *c,
+vec4_visitor::vec4_visitor(const struct brw_compiler *compiler,
+ void *log_data,
struct gl_program *prog,
const struct brw_vue_prog_key *key,
struct brw_vue_prog_data *prog_data,
gl_shader_stage stage,
void *mem_ctx,
bool no_spills,
- shader_time_shader_type st_base,
- shader_time_shader_type st_written,
- shader_time_shader_type st_reset)
- : backend_visitor(brw, shader_prog, prog, &prog_data->base, stage),
- c(c),
+ int shader_time_index)
+ : backend_shader(compiler, log_data, mem_ctx,
+ shader_prog, prog, &prog_data->base, stage),
key(key),
prog_data(prog_data),
sanity_param_count(0),
first_non_payload_grf(0),
need_all_constants_in_pull_buffer(false),
no_spills(no_spills),
- st_base(st_base),
- st_written(st_written),
- st_reset(st_reset)
+ shader_time_index(shader_time_index),
+ last_scratch(0)
{
- this->mem_ctx = mem_ctx;
this->failed = false;
this->base_ir = NULL;
this->virtual_grf_end = NULL;
this->live_intervals = NULL;
- this->max_grf = brw->gen >= 7 ? GEN7_MRF_HACK_START : BRW_MAX_GRF;
+ this->max_grf = devinfo->gen >= 7 ? GEN7_MRF_HACK_START : BRW_MAX_GRF;
this->uniforms = 0;