* IN THE SOFTWARE.
*/
-#include "glsl/ir.h"
-#include "glsl/ir_optimization.h"
-#include "glsl/nir/glsl_to_nir.h"
+#include "compiler/glsl/ir.h"
#include "main/shaderimage.h"
-#include "program/prog_to_nir.h"
#include "brw_fs.h"
#include "brw_fs_surface_builder.h"
-#include "brw_vec4_gs_visitor.h"
#include "brw_nir.h"
-#include "brw_fs_surface_builder.h"
-#include "brw_vec4_gs_visitor.h"
+#include "brw_program.h"
using namespace brw;
using namespace brw::surface_access;
nir_emit_system_values();
/* get the main function and emit it */
- nir_foreach_overload(nir, overload) {
- assert(strcmp(overload->function->name, "main") == 0);
- assert(overload->impl);
- nir_emit_impl(overload->impl);
+ nir_foreach_function(nir, function) {
+ assert(strcmp(function->name, "main") == 0);
+ assert(function->impl);
+ nir_emit_impl(function->impl);
}
}
reg.type = BRW_REGISTER_TYPE_D;
bld.emit(FS_OPCODE_CINTERP, retype(input, BRW_REGISTER_TYPE_D), reg);
} else {
- emit_general_interpolation(input, var->name, var->type,
+ int location = var->data.location;
+ emit_general_interpolation(&input, var->name, var->type,
(glsl_interp_qualifier) var->data.interpolation,
- var->data.location, var->data.centroid,
+ &location, var->data.centroid,
var->data.sample);
}
}
}
+void
+fs_visitor::nir_setup_single_output_varying(fs_reg *reg,
+ const glsl_type *type,
+ unsigned *location)
+{
+ if (type->is_array() || type->is_matrix()) {
+ const struct glsl_type *elem_type = glsl_get_array_element(type);
+ const unsigned length = glsl_get_length(type);
+
+ for (unsigned i = 0; i < length; i++) {
+ nir_setup_single_output_varying(reg, elem_type, location);
+ }
+ } else if (type->is_record()) {
+ for (unsigned i = 0; i < type->length; i++) {
+ const struct glsl_type *field_type = type->fields.structure[i].type;
+ nir_setup_single_output_varying(reg, field_type, location);
+ }
+ } else {
+ assert(type->is_scalar() || type->is_vector());
+ this->outputs[*location] = *reg;
+ this->output_components[*location] = type->vector_elements;
+ *reg = offset(*reg, bld, 4);
+ (*location)++;
+ }
+}
+
void
fs_visitor::nir_setup_outputs()
{
nir_foreach_variable(var, &nir->outputs) {
fs_reg reg = offset(nir_outputs, bld, var->data.driver_location);
- int vector_elements = var->type->without_array()->vector_elements;
-
switch (stage) {
case MESA_SHADER_VERTEX:
- case MESA_SHADER_GEOMETRY:
- for (int i = 0; i < type_size_vec4(var->type); i++) {
- int output = var->data.location + i;
- this->outputs[output] = offset(reg, bld, 4 * i);
- this->output_components[output] = vector_elements;
- }
+ case MESA_SHADER_TESS_EVAL:
+ case MESA_SHADER_GEOMETRY: {
+ unsigned location = var->data.location;
+ nir_setup_single_output_varying(®, var->type, &location);
break;
+ }
case MESA_SHADER_FRAGMENT:
- if (var->data.index > 0) {
+ if (key->force_dual_color_blend &&
+ var->data.location == FRAG_RESULT_DATA1) {
+ this->dual_src_output = reg;
+ this->do_dual_src = true;
+ } else if (var->data.index > 0) {
assert(var->data.location == FRAG_RESULT_DATA0);
assert(var->data.index == 1);
this->dual_src_output = reg;
} else if (var->data.location == FRAG_RESULT_SAMPLE_MASK) {
this->sample_mask = reg;
} else {
+ int vector_elements = var->type->without_array()->vector_elements;
+
/* gl_FragData or a user-defined FS output */
assert(var->data.location >= FRAG_RESULT_DATA0 &&
var->data.location < FRAG_RESULT_DATA0+BRW_MAX_DRAW_BUFFERS);
if (dispatch_width != 8)
return;
- uniforms = nir->num_uniforms;
-
- nir_foreach_variable(var, &nir->uniforms) {
- /* UBO's and atomics don't take up space in the uniform file */
- if (var->interface_type != NULL || var->type->contains_atomic())
- continue;
-
- if (type_size_scalar(var->type) > 0)
- param_size[var->data.driver_location] = type_size_scalar(var->type);
- }
+ uniforms = nir->num_uniforms / 4;
}
static bool
*reg = *v->emit_vs_system_value(SYSTEM_VALUE_INSTANCE_ID);
break;
+ case nir_intrinsic_load_base_instance:
+ assert(v->stage == MESA_SHADER_VERTEX);
+ reg = &v->nir_system_values[SYSTEM_VALUE_BASE_INSTANCE];
+ if (reg->file == BAD_FILE)
+ *reg = *v->emit_vs_system_value(SYSTEM_VALUE_BASE_INSTANCE);
+ break;
+
+ case nir_intrinsic_load_draw_id:
+ assert(v->stage == MESA_SHADER_VERTEX);
+ reg = &v->nir_system_values[SYSTEM_VALUE_DRAW_ID];
+ if (reg->file == BAD_FILE)
+ *reg = *v->emit_vs_system_value(SYSTEM_VALUE_DRAW_ID);
+ break;
+
case nir_intrinsic_load_invocation_id:
assert(v->stage == MESA_SHADER_GEOMETRY);
reg = &v->nir_system_values[SYSTEM_VALUE_INVOCATION_ID];
const fs_builder abld = v->bld.annotate("gl_InvocationID", NULL);
fs_reg g1(retype(brw_vec8_grf(1, 0), BRW_REGISTER_TYPE_UD));
fs_reg iid = abld.vgrf(BRW_REGISTER_TYPE_UD, 1);
- abld.SHR(iid, g1, fs_reg(27u));
+ abld.SHR(iid, g1, brw_imm_ud(27u));
*reg = iid;
}
break;
*reg = *v->emit_cs_work_group_id_setup();
break;
+ case nir_intrinsic_load_helper_invocation:
+ assert(v->stage == MESA_SHADER_FRAGMENT);
+ reg = &v->nir_system_values[SYSTEM_VALUE_HELPER_INVOCATION];
+ if (reg->file == BAD_FILE) {
+ const fs_builder abld =
+ v->bld.annotate("gl_HelperInvocation", NULL);
+
+ /* On Gen6+ (gl_HelperInvocation is only exposed on Gen7+) the
+ * pixel mask is in g1.7 of the thread payload.
+ *
+ * We move the per-channel pixel enable bit to the low bit of each
+ * channel by shifting the byte containing the pixel mask by the
+ * vector immediate 0x76543210UV.
+ *
+ * The region of <1,8,0> reads only 1 byte (the pixel masks for
+ * subspans 0 and 1) in SIMD8 and an additional byte (the pixel
+ * masks for 2 and 3) in SIMD16.
+ */
+ fs_reg shifted = abld.vgrf(BRW_REGISTER_TYPE_UW, 1);
+ abld.SHR(shifted,
+ stride(byte_offset(retype(brw_vec1_grf(1, 0),
+ BRW_REGISTER_TYPE_UB), 28),
+ 1, 8, 0),
+ brw_imm_uv(0x76543210));
+
+ /* A set bit in the pixel mask means the channel is enabled, but
+ * that is the opposite of gl_HelperInvocation so we need to invert
+ * the mask.
+ *
+ * The negate source-modifier bit of logical instructions on Gen8+
+ * performs 1's complement negation, so we can use that instead of
+ * a NOT instruction.
+ */
+ fs_reg inverted = negate(shifted);
+ if (v->devinfo->gen < 8) {
+ inverted = abld.vgrf(BRW_REGISTER_TYPE_UW);
+ abld.NOT(inverted, shifted);
+ }
+
+ /* We then resolve the 0/1 result to 0/~0 boolean values by ANDing
+ * with 1 and negating.
+ */
+ fs_reg anded = abld.vgrf(BRW_REGISTER_TYPE_UD, 1);
+ abld.AND(anded, inverted, brw_imm_uw(1));
+
+ fs_reg dst = abld.vgrf(BRW_REGISTER_TYPE_D, 1);
+ abld.MOV(dst, negate(retype(anded, BRW_REGISTER_TYPE_D)));
+ *reg = dst;
+ }
+ break;
+
default:
break;
}
nir_system_values[i] = fs_reg();
}
- nir_foreach_overload(nir, overload) {
- assert(strcmp(overload->function->name, "main") == 0);
- assert(overload->impl);
- nir_foreach_block(overload->impl, emit_system_values_block, this);
+ nir_foreach_function(nir, function) {
+ assert(strcmp(function->name, "main") == 0);
+ assert(function->impl);
+ nir_foreach_block(function->impl, emit_system_values_block, this);
}
}
case MESA_SHADER_VERTEX:
nir_emit_vs_intrinsic(abld, nir_instr_as_intrinsic(instr));
break;
+ case MESA_SHADER_TESS_EVAL:
+ nir_emit_tes_intrinsic(abld, nir_instr_as_intrinsic(instr));
+ break;
case MESA_SHADER_GEOMETRY:
nir_emit_gs_intrinsic(abld, nir_instr_as_intrinsic(instr));
break;
}
}
+/**
+ * Recognizes a parent instruction of nir_op_extract_* and changes the type to
+ * match instr.
+ */
+bool
+fs_visitor::optimize_extract_to_float(nir_alu_instr *instr,
+ const fs_reg &result)
+{
+ if (!instr->src[0].src.is_ssa ||
+ !instr->src[0].src.ssa->parent_instr)
+ return false;
+
+ if (instr->src[0].src.ssa->parent_instr->type != nir_instr_type_alu)
+ return false;
+
+ nir_alu_instr *src0 =
+ nir_instr_as_alu(instr->src[0].src.ssa->parent_instr);
+
+ if (src0->op != nir_op_extract_u8 && src0->op != nir_op_extract_u16 &&
+ src0->op != nir_op_extract_i8 && src0->op != nir_op_extract_i16)
+ return false;
+
+ nir_const_value *element = nir_src_as_const_value(src0->src[1].src);
+ assert(element != NULL);
+
+ enum opcode extract_op;
+ if (src0->op == nir_op_extract_u16 || src0->op == nir_op_extract_i16) {
+ assert(element->u32[0] <= 1);
+ extract_op = SHADER_OPCODE_EXTRACT_WORD;
+ } else {
+ assert(element->u32[0] <= 3);
+ extract_op = SHADER_OPCODE_EXTRACT_BYTE;
+ }
+
+ fs_reg op0 = get_nir_src(src0->src[0].src);
+ op0.type = brw_type_for_nir_type(nir_op_infos[src0->op].input_types[0]);
+ op0 = offset(op0, bld, src0->src[0].swizzle[0]);
+
+ set_saturate(instr->dest.saturate,
+ bld.emit(extract_op, result, op0, brw_imm_ud(element->u32[0])));
+ return true;
+}
+
bool
fs_visitor::optimize_frontfacing_ternary(nir_alu_instr *instr,
const fs_reg &result)
return false;
nir_const_value *value1 = nir_src_as_const_value(instr->src[1].src);
- if (!value1 || fabsf(value1->f[0]) != 1.0f)
+ if (!value1 || fabsf(value1->f32[0]) != 1.0f)
return false;
nir_const_value *value2 = nir_src_as_const_value(instr->src[2].src);
- if (!value2 || fabsf(value2->f[0]) != 1.0f)
+ if (!value2 || fabsf(value2->f32[0]) != 1.0f)
return false;
fs_reg tmp = vgrf(glsl_type::int_type);
* surely be TRIANGLES
*/
- if (value1->f[0] == -1.0f) {
+ if (value1->f32[0] == -1.0f) {
g0.negate = true;
}
tmp.subreg_offset = 2;
tmp.stride = 2;
- fs_inst *or_inst = bld.OR(tmp, g0, fs_reg(0x3f80));
- or_inst->src[1].type = BRW_REGISTER_TYPE_UW;
+ bld.OR(tmp, g0, brw_imm_uw(0x3f80));
tmp.type = BRW_REGISTER_TYPE_D;
tmp.subreg_offset = 0;
* surely be TRIANGLES
*/
- if (value1->f[0] == -1.0f) {
+ if (value1->f32[0] == -1.0f) {
g1_6.negate = true;
}
- bld.OR(tmp, g1_6, fs_reg(0x3f800000));
+ bld.OR(tmp, g1_6, brw_imm_d(0x3f800000));
}
- bld.AND(retype(result, BRW_REGISTER_TYPE_D), tmp, fs_reg(0xbf800000));
+ bld.AND(retype(result, BRW_REGISTER_TYPE_D), tmp, brw_imm_d(0xbf800000));
return true;
}
switch (instr->op) {
case nir_op_i2f:
case nir_op_u2f:
+ if (optimize_extract_to_float(instr, result))
+ return;
+
inst = bld.MOV(result, op[0]);
inst->saturate = instr->dest.saturate;
break;
* Predicated OR ORs 1.0 (0x3f800000) with the sign bit if val is not
* zero.
*/
- bld.CMP(bld.null_reg_f(), op[0], fs_reg(0.0f), BRW_CONDITIONAL_NZ);
+ bld.CMP(bld.null_reg_f(), op[0], brw_imm_f(0.0f), BRW_CONDITIONAL_NZ);
fs_reg result_int = retype(result, BRW_REGISTER_TYPE_UD);
op[0].type = BRW_REGISTER_TYPE_UD;
result.type = BRW_REGISTER_TYPE_UD;
- bld.AND(result_int, op[0], fs_reg(0x80000000u));
+ bld.AND(result_int, op[0], brw_imm_ud(0x80000000u));
- inst = bld.OR(result_int, result_int, fs_reg(0x3f800000u));
+ inst = bld.OR(result_int, result_int, brw_imm_ud(0x3f800000u));
inst->predicate = BRW_PREDICATE_NORMAL;
if (instr->dest.saturate) {
inst = bld.MOV(result, result);
* -> non-negative val generates 0x00000000.
* Predicated OR sets 1 if val is positive.
*/
- bld.CMP(bld.null_reg_d(), op[0], fs_reg(0), BRW_CONDITIONAL_G);
- bld.ASR(result, op[0], fs_reg(31));
- inst = bld.OR(result, result, fs_reg(1));
+ bld.CMP(bld.null_reg_d(), op[0], brw_imm_d(0), BRW_CONDITIONAL_G);
+ bld.ASR(result, op[0], brw_imm_d(31));
+ inst = bld.OR(result, result, brw_imm_d(1));
inst->predicate = BRW_PREDICATE_NORMAL;
break;
case nir_op_fddy:
if (fs_key->high_quality_derivatives) {
inst = bld.emit(FS_OPCODE_DDY_FINE, result, op[0],
- fs_reg(fs_key->render_to_fbo));
+ brw_imm_d(fs_key->render_to_fbo));
} else {
inst = bld.emit(FS_OPCODE_DDY_COARSE, result, op[0],
- fs_reg(fs_key->render_to_fbo));
+ brw_imm_d(fs_key->render_to_fbo));
}
inst->saturate = instr->dest.saturate;
break;
case nir_op_fddy_fine:
inst = bld.emit(FS_OPCODE_DDY_FINE, result, op[0],
- fs_reg(fs_key->render_to_fbo));
+ brw_imm_d(fs_key->render_to_fbo));
inst->saturate = instr->dest.saturate;
break;
case nir_op_fddy_coarse:
inst = bld.emit(FS_OPCODE_DDY_COARSE, result, op[0],
- fs_reg(fs_key->render_to_fbo));
+ brw_imm_d(fs_key->render_to_fbo));
inst->saturate = instr->dest.saturate;
break;
unreachable("Should have been lowered by borrow_to_arith().");
case nir_op_umod:
+ case nir_op_irem:
+ /* According to the sign table for INT DIV in the Ivy Bridge PRM, it
+ * appears that our hardware just does the right thing for signed
+ * remainder.
+ */
+ bld.emit(SHADER_OPCODE_INT_REMAINDER, result, op[0], op[1]);
+ break;
+
+ case nir_op_imod: {
+ /* Get a regular C-style remainder. If a % b == 0, set the predicate. */
bld.emit(SHADER_OPCODE_INT_REMAINDER, result, op[0], op[1]);
+
+ /* Math instructions don't support conditional mod */
+ inst = bld.MOV(bld.null_reg_d(), result);
+ inst->conditional_mod = BRW_CONDITIONAL_NZ;
+
+ /* Now, we need to determine if signs of the sources are different.
+ * When we XOR the sources, the top bit is 0 if they are the same and 1
+ * if they are different. We can then use a conditional modifier to
+ * turn that into a predicate. This leads us to an XOR.l instruction.
+ *
+ * Technically, according to the PRM, you're not allowed to use .l on a
+ * XOR instruction. However, emperical experiments and Curro's reading
+ * of the simulator source both indicate that it's safe.
+ */
+ fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_D);
+ inst = bld.XOR(tmp, op[0], op[1]);
+ inst->predicate = BRW_PREDICATE_NORMAL;
+ inst->conditional_mod = BRW_CONDITIONAL_L;
+
+ /* If the result of the initial remainder operation is non-zero and the
+ * two sources have different signs, add in a copy of op[1] to get the
+ * final integer modulus value.
+ */
+ inst = bld.ADD(result, result, op[1]);
+ inst->predicate = BRW_PREDICATE_NORMAL;
break;
+ }
case nir_op_flt:
case nir_op_ilt:
case nir_op_fdot2:
case nir_op_fdot3:
case nir_op_fdot4:
- case nir_op_bany2:
- case nir_op_bany3:
- case nir_op_bany4:
- case nir_op_ball2:
- case nir_op_ball3:
- case nir_op_ball4:
case nir_op_ball_fequal2:
case nir_op_ball_iequal2:
case nir_op_ball_fequal3:
break;
case nir_op_f2b:
- bld.CMP(result, op[0], fs_reg(0.0f), BRW_CONDITIONAL_NZ);
+ bld.CMP(result, op[0], brw_imm_f(0.0f), BRW_CONDITIONAL_NZ);
break;
case nir_op_i2b:
- bld.CMP(result, op[0], fs_reg(0), BRW_CONDITIONAL_NZ);
+ bld.CMP(result, op[0], brw_imm_d(0), BRW_CONDITIONAL_NZ);
break;
case nir_op_ftrunc:
inst->saturate = instr->dest.saturate;
break;
+ case nir_op_fquantize2f16: {
+ fs_reg tmp16 = bld.vgrf(BRW_REGISTER_TYPE_D);
+ fs_reg tmp32 = bld.vgrf(BRW_REGISTER_TYPE_F);
+ fs_reg zero = bld.vgrf(BRW_REGISTER_TYPE_F);
+
+ /* The destination stride must be at least as big as the source stride. */
+ tmp16.type = BRW_REGISTER_TYPE_W;
+ tmp16.stride = 2;
+
+ /* Check for denormal */
+ fs_reg abs_src0 = op[0];
+ abs_src0.abs = true;
+ bld.CMP(bld.null_reg_f(), abs_src0, brw_imm_f(ldexpf(1.0, -14)),
+ BRW_CONDITIONAL_L);
+ /* Get the appropriately signed zero */
+ bld.AND(retype(zero, BRW_REGISTER_TYPE_UD),
+ retype(op[0], BRW_REGISTER_TYPE_UD),
+ brw_imm_ud(0x80000000));
+ /* Do the actual F32 -> F16 -> F32 conversion */
+ bld.emit(BRW_OPCODE_F32TO16, tmp16, op[0]);
+ bld.emit(BRW_OPCODE_F16TO32, tmp32, tmp16);
+ /* Select that or zero based on normal status */
+ inst = bld.SEL(result, zero, tmp32);
+ inst->predicate = BRW_PREDICATE_NORMAL;
+ inst->saturate = instr->dest.saturate;
+ break;
+ }
+
case nir_op_fmin:
case nir_op_imin:
case nir_op_umin:
- if (devinfo->gen >= 6) {
- inst = bld.emit(BRW_OPCODE_SEL, result, op[0], op[1]);
- inst->conditional_mod = BRW_CONDITIONAL_L;
- } else {
- bld.CMP(bld.null_reg_d(), op[0], op[1], BRW_CONDITIONAL_L);
- inst = bld.SEL(result, op[0], op[1]);
- inst->predicate = BRW_PREDICATE_NORMAL;
- }
+ inst = bld.emit_minmax(result, op[0], op[1], BRW_CONDITIONAL_L);
inst->saturate = instr->dest.saturate;
break;
case nir_op_fmax:
case nir_op_imax:
case nir_op_umax:
- if (devinfo->gen >= 6) {
- inst = bld.emit(BRW_OPCODE_SEL, result, op[0], op[1]);
- inst->conditional_mod = BRW_CONDITIONAL_GE;
- } else {
- bld.CMP(bld.null_reg_d(), op[0], op[1], BRW_CONDITIONAL_GE);
- inst = bld.SEL(result, op[0], op[1]);
- inst->predicate = BRW_PREDICATE_NORMAL;
- }
+ inst = bld.emit_minmax(result, op[0], op[1], BRW_CONDITIONAL_GE);
inst->saturate = instr->dest.saturate;
break;
* from the LSB side. If FBH didn't return an error (0xFFFFFFFF), then
* subtract the result from 31 to convert the MSB count into an LSB count.
*/
- bld.CMP(bld.null_reg_d(), result, fs_reg(-1), BRW_CONDITIONAL_NZ);
+ bld.CMP(bld.null_reg_d(), result, brw_imm_d(-1), BRW_CONDITIONAL_NZ);
- inst = bld.ADD(result, result, fs_reg(31));
+ inst = bld.ADD(result, result, brw_imm_d(31));
inst->predicate = BRW_PREDICATE_NORMAL;
inst->src[0].negate = true;
break;
case nir_op_ubitfield_extract:
case nir_op_ibitfield_extract:
+ unreachable("should have been lowered");
+ case nir_op_ubfe:
+ case nir_op_ibfe:
bld.BFE(result, op[2], op[1], op[0]);
break;
case nir_op_bfm:
break;
case nir_op_bitfield_insert:
- unreachable("not reached: should be handled by "
- "lower_instructions::bitfield_insert_to_bfm_bfi");
+ unreachable("not reached: should have been lowered");
case nir_op_ishl:
bld.SHL(result, op[0], op[1]);
if (optimize_frontfacing_ternary(instr, result))
return;
- bld.CMP(bld.null_reg_d(), op[0], fs_reg(0), BRW_CONDITIONAL_NZ);
+ bld.CMP(bld.null_reg_d(), op[0], brw_imm_d(0), BRW_CONDITIONAL_NZ);
inst = bld.SEL(result, op[1], op[2]);
inst->predicate = BRW_PREDICATE_NORMAL;
break;
+ case nir_op_extract_u8:
+ case nir_op_extract_i8: {
+ nir_const_value *byte = nir_src_as_const_value(instr->src[1].src);
+ bld.emit(SHADER_OPCODE_EXTRACT_BYTE,
+ result, op[0], brw_imm_ud(byte->u32[0]));
+ break;
+ }
+
+ case nir_op_extract_u16:
+ case nir_op_extract_i16: {
+ nir_const_value *word = nir_src_as_const_value(instr->src[1].src);
+ bld.emit(SHADER_OPCODE_EXTRACT_WORD,
+ result, op[0], brw_imm_ud(word->u32[0]));
+ break;
+ }
+
default:
unreachable("unhandled instruction");
}
if (devinfo->gen <= 5 &&
(instr->instr.pass_flags & BRW_NIR_BOOLEAN_MASK) == BRW_NIR_BOOLEAN_NEEDS_RESOLVE) {
fs_reg masked = vgrf(glsl_type::int_type);
- bld.AND(masked, result, fs_reg(1));
+ bld.AND(masked, result, brw_imm_d(1));
masked.negate = true;
bld.MOV(retype(result, BRW_REGISTER_TYPE_D), masked);
}
fs_reg reg = bld.vgrf(BRW_REGISTER_TYPE_D, instr->def.num_components);
for (unsigned i = 0; i < instr->def.num_components; i++)
- bld.MOV(offset(reg, bld, i), fs_reg(instr->value.i[i]));
+ bld.MOV(offset(reg, bld, i), brw_imm_d(instr->value.i32[i]));
nir_ssa_values[instr->def.index] = reg;
}
instr->def.num_components);
}
-static fs_reg
-fs_reg_for_nir_reg(fs_visitor *v, nir_register *nir_reg,
- unsigned base_offset, nir_src *indirect)
-{
- fs_reg reg;
-
- assert(!nir_reg->is_global);
-
- reg = v->nir_locals[nir_reg->index];
-
- reg = offset(reg, v->bld, base_offset * nir_reg->num_components);
- if (indirect) {
- int multiplier = nir_reg->num_components * (v->dispatch_width / 8);
-
- reg.reladdr = new(v->mem_ctx) fs_reg(v->vgrf(glsl_type::int_type));
- v->bld.MUL(*reg.reladdr, v->get_nir_src(*indirect),
- fs_reg(multiplier));
- }
-
- return reg;
-}
-
fs_reg
fs_visitor::get_nir_src(nir_src src)
{
if (src.is_ssa) {
reg = nir_ssa_values[src.ssa->index];
} else {
- reg = fs_reg_for_nir_reg(this, src.reg.reg, src.reg.base_offset,
- src.reg.indirect);
+ /* We don't handle indirects on locals */
+ assert(src.reg.indirect == NULL);
+ reg = offset(nir_locals[src.reg.reg->index], bld,
+ src.reg.base_offset * src.reg.reg->num_components);
}
/* to avoid floating-point denorm flushing problems, set the type by
nir_ssa_values[dest.ssa.index] = bld.vgrf(BRW_REGISTER_TYPE_F,
dest.ssa.num_components);
return nir_ssa_values[dest.ssa.index];
+ } else {
+ /* We don't handle indirects on locals */
+ assert(dest.reg.indirect == NULL);
+ return offset(nir_locals[dest.reg.reg->index], bld,
+ dest.reg.base_offset * dest.reg.reg->num_components);
}
-
- return fs_reg_for_nir_reg(this, dest.reg.reg, dest.reg.base_offset,
- dest.reg.indirect);
}
fs_reg
fs_visitor::get_nir_image_deref(const nir_deref_var *deref)
{
- fs_reg image(UNIFORM, deref->var->data.driver_location,
+ fs_reg image(UNIFORM, deref->var->data.driver_location / 4,
BRW_REGISTER_TYPE_UD);
+ fs_reg indirect;
+ unsigned indirect_max = 0;
for (const nir_deref *tail = &deref->deref; tail->child;
tail = tail->child) {
image = offset(image, bld, base * element_size);
if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
- fs_reg tmp = vgrf(glsl_type::int_type);
+ fs_reg tmp = vgrf(glsl_type::uint_type);
if (devinfo->gen == 7 && !devinfo->is_haswell) {
/* IVB hangs when trying to access an invalid surface index with
*/
bld.emit_minmax(tmp, retype(get_nir_src(deref_array->indirect),
BRW_REGISTER_TYPE_UD),
- fs_reg(size - base - 1), BRW_CONDITIONAL_L);
+ brw_imm_ud(size - base - 1), BRW_CONDITIONAL_L);
} else {
bld.MOV(tmp, get_nir_src(deref_array->indirect));
}
- bld.MUL(tmp, tmp, fs_reg(element_size));
- if (image.reladdr)
- bld.ADD(*image.reladdr, *image.reladdr, tmp);
- else
- image.reladdr = new(mem_ctx) fs_reg(tmp);
+ indirect_max += element_size * (tail->type->length - 1);
+
+ bld.MUL(tmp, tmp, brw_imm_ud(element_size * 4));
+ if (indirect.file == BAD_FILE) {
+ indirect = tmp;
+ } else {
+ bld.ADD(indirect, indirect, tmp);
+ }
}
}
- return image;
+ if (indirect.file == BAD_FILE) {
+ return image;
+ } else {
+ /* Emit a pile of MOVs to load the uniform into a temporary. The
+ * dead-code elimination pass will get rid of what we don't use.
+ */
+ fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD, BRW_IMAGE_PARAM_SIZE);
+ for (unsigned j = 0; j < BRW_IMAGE_PARAM_SIZE; j++) {
+ bld.emit(SHADER_OPCODE_MOV_INDIRECT,
+ offset(tmp, bld, j), offset(image, bld, j),
+ indirect, brw_imm_ud((indirect_max + 1) * 4));
+ }
+ return tmp;
+ }
}
void
fs_inst *new_inst = new(mem_ctx) fs_inst(inst);
new_inst->dst = offset(new_inst->dst, bld, i);
for (unsigned j = 0; j < new_inst->sources; j++)
- if (new_inst->src[j].file == GRF)
+ if (new_inst->src[j].file == VGRF)
new_inst->src[j] = offset(new_inst->src[j], bld, i);
bld.emit(new_inst);
static brw_reg_type
get_image_base_type(const glsl_type *type)
{
- switch ((glsl_base_type)type->sampler_type) {
+ switch ((glsl_base_type)type->sampled_type) {
case GLSL_TYPE_UINT:
return BRW_REGISTER_TYPE_UD;
case GLSL_TYPE_INT:
fs_reg result = bld.vgrf(x.type, 1);
fs_reg one = bld.vgrf(x.type, 1);
- bld.MOV(one, retype(fs_reg(1), one.type));
+ bld.MOV(one, retype(brw_imm_d(1), one.type));
bld.SHL(result, one, x);
return result;
}
/* control_data_bits |= 1 << ((vertex_count - 1) % 32) */
fs_reg prev_count = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
- abld.ADD(prev_count, vertex_count, fs_reg(0xffffffffu));
+ abld.ADD(prev_count, vertex_count, brw_imm_ud(0xffffffffu));
fs_reg mask = intexp2(abld, prev_count);
/* Note: we're relying on the fact that the GEN SHL instruction only pays
* attention to the lower 5 bits of its second source argument, so on this
if (opcode != SHADER_OPCODE_URB_WRITE_SIMD8) {
fs_reg dword_index = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
fs_reg prev_count = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
- abld.ADD(prev_count, vertex_count, fs_reg(0xffffffffu));
+ abld.ADD(prev_count, vertex_count, brw_imm_ud(0xffffffffu));
unsigned log2_bits_per_vertex =
_mesa_fls(gs_compile->control_data_bits_per_vertex);
- abld.SHR(dword_index, prev_count, fs_reg(6u - log2_bits_per_vertex));
+ abld.SHR(dword_index, prev_count, brw_imm_ud(6u - log2_bits_per_vertex));
if (per_slot_offset.file != BAD_FILE) {
/* Set the per-slot offset to dword_index / 4, so that we'll write to
* the appropriate OWord within the control data header.
*/
- abld.SHR(per_slot_offset, dword_index, fs_reg(2u));
+ abld.SHR(per_slot_offset, dword_index, brw_imm_ud(2u));
}
/* Set the channel masks to 1 << (dword_index % 4), so that we'll
* write to the appropriate DWORD within the OWORD.
*/
fs_reg channel = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
- fwa_bld.AND(channel, dword_index, fs_reg(3u));
+ fwa_bld.AND(channel, dword_index, brw_imm_ud(3u));
channel_mask = intexp2(fwa_bld, channel);
/* Then the channel masks need to be in bits 23:16. */
- fwa_bld.SHL(channel_mask, channel_mask, fs_reg(16u));
+ fwa_bld.SHL(channel_mask, channel_mask, brw_imm_ud(16u));
}
/* Store the control data bits in the message payload and send it. */
/* reg::sid = stream_id */
fs_reg sid = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
- abld.MOV(sid, fs_reg(stream_id));
+ abld.MOV(sid, brw_imm_ud(stream_id));
/* reg:shift_count = 2 * (vertex_count - 1) */
fs_reg shift_count = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
- abld.SHL(shift_count, vertex_count, fs_reg(1u));
+ abld.SHL(shift_count, vertex_count, brw_imm_ud(1u));
/* Note: we're relying on the fact that the GEN SHL instruction only pays
* attention to the lower 5 bits of its second source argument, so on this
*/
fs_inst *inst =
abld.AND(bld.null_reg_d(), vertex_count,
- fs_reg(32u / gs_compile->control_data_bits_per_vertex - 1u));
+ brw_imm_ud(32u / gs_compile->control_data_bits_per_vertex - 1u));
inst->conditional_mod = BRW_CONDITIONAL_Z;
abld.IF(BRW_PREDICATE_NORMAL);
/* If vertex_count is 0, then no control data bits have been
* accumulated yet, so we can skip emitting them.
*/
- abld.CMP(bld.null_reg_d(), vertex_count, fs_reg(0u),
+ abld.CMP(bld.null_reg_d(), vertex_count, brw_imm_ud(0u),
BRW_CONDITIONAL_NEQ);
abld.IF(BRW_PREDICATE_NORMAL);
emit_gs_control_data_bits(vertex_count);
* effect of any call to EndPrimitive() that the shader may have
* made before outputting its first vertex.
*/
- inst = abld.MOV(this->control_data_bits, fs_reg(0u));
+ inst = abld.MOV(this->control_data_bits, brw_imm_ud(0u));
inst->force_writemask_all = true;
abld.emit(BRW_OPCODE_ENDIF);
}
void
fs_visitor::emit_gs_input_load(const fs_reg &dst,
const nir_src &vertex_src,
- unsigned input_offset,
+ unsigned base_offset,
+ const nir_src &offset_src,
unsigned num_components)
{
- const brw_vue_prog_data *vue_prog_data = (const brw_vue_prog_data *) prog_data;
- const unsigned vertex = nir_src_as_const_value(vertex_src)->u[0];
+ struct brw_gs_prog_data *gs_prog_data = (struct brw_gs_prog_data *) prog_data;
- const unsigned array_stride = vue_prog_data->urb_read_length * 8;
+ nir_const_value *vertex_const = nir_src_as_const_value(vertex_src);
+ nir_const_value *offset_const = nir_src_as_const_value(offset_src);
+ const unsigned push_reg_count = gs_prog_data->base.urb_read_length * 8;
- const bool pushed = 4 * input_offset < array_stride;
-
- if (input_offset == 0) {
- /* This is the VUE header, containing VARYING_SLOT_LAYER [.y],
- * VARYING_SLOT_VIEWPORT [.z], and VARYING_SLOT_PSIZ [.w].
- * Only gl_PointSize is available as a GS input, so they must
- * be asking for that input.
- */
- if (pushed) {
- bld.MOV(dst, fs_reg(ATTR, array_stride * vertex + 3, dst.type));
+ /* Offset 0 is the VUE header, which contains VARYING_SLOT_LAYER [.y],
+ * VARYING_SLOT_VIEWPORT [.z], and VARYING_SLOT_PSIZ [.w]. Only
+ * gl_PointSize is available as a GS input, however, so it must be that.
+ */
+ const bool is_point_size = (base_offset == 0);
+
+ if (offset_const != NULL && vertex_const != NULL &&
+ 4 * (base_offset + offset_const->u32[0]) < push_reg_count) {
+ int imm_offset = (base_offset + offset_const->u32[0]) * 4 +
+ vertex_const->u32[0] * push_reg_count;
+ /* This input was pushed into registers. */
+ if (is_point_size) {
+ /* gl_PointSize comes in .w */
+ assert(imm_offset == 0);
+ bld.MOV(dst, fs_reg(ATTR, imm_offset + 3, dst.type));
} else {
- fs_reg tmp = bld.vgrf(dst.type, 4);
- fs_inst *inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp,
- fs_reg(vertex), fs_reg(0));
- inst->regs_written = 4;
- bld.MOV(dst, offset(tmp, bld, 3));
- }
- } else {
- if (pushed) {
- int index = vertex * array_stride + 4 * input_offset;
for (unsigned i = 0; i < num_components; i++) {
- bld.MOV(offset(dst, bld, i), fs_reg(ATTR, index + i, dst.type));
+ bld.MOV(offset(dst, bld, i),
+ fs_reg(ATTR, imm_offset + i, dst.type));
}
+ }
+ } else {
+ /* Resort to the pull model. Ensure the VUE handles are provided. */
+ gs_prog_data->base.include_vue_handles = true;
+
+ unsigned first_icp_handle = gs_prog_data->include_primitive_id ? 3 : 2;
+ fs_reg icp_handle;
+
+ if (vertex_const) {
+ /* The vertex index is constant; just select the proper URB handle. */
+ icp_handle =
+ retype(brw_vec8_grf(first_icp_handle + vertex_const->i32[0], 0),
+ BRW_REGISTER_TYPE_UD);
+ } else {
+ /* The vertex index is non-constant. We need to use indirect
+ * addressing to fetch the proper URB handle.
+ *
+ * First, we start with the sequence <7, 6, 5, 4, 3, 2, 1, 0>
+ * indicating that channel <n> should read the handle from
+ * DWord <n>. We convert that to bytes by multiplying by 4.
+ *
+ * Next, we convert the vertex index to bytes by multiplying
+ * by 32 (shifting by 5), and add the two together. This is
+ * the final indirect byte offset.
+ */
+ fs_reg sequence = bld.vgrf(BRW_REGISTER_TYPE_W, 1);
+ fs_reg channel_offsets = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
+ fs_reg vertex_offset_bytes = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
+ fs_reg icp_offset_bytes = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
+ icp_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
+
+ /* sequence = <7, 6, 5, 4, 3, 2, 1, 0> */
+ bld.MOV(sequence, fs_reg(brw_imm_v(0x76543210)));
+ /* channel_offsets = 4 * sequence = <28, 24, 20, 16, 12, 8, 4, 0> */
+ bld.SHL(channel_offsets, sequence, brw_imm_ud(2u));
+ /* Convert vertex_index to bytes (multiply by 32) */
+ bld.SHL(vertex_offset_bytes,
+ retype(get_nir_src(vertex_src), BRW_REGISTER_TYPE_UD),
+ brw_imm_ud(5u));
+ bld.ADD(icp_offset_bytes, vertex_offset_bytes, channel_offsets);
+
+ /* Use first_icp_handle as the base offset. There is one register
+ * of URB handles per vertex, so inform the register allocator that
+ * we might read up to nir->info.gs.vertices_in registers.
+ */
+ bld.emit(SHADER_OPCODE_MOV_INDIRECT, icp_handle,
+ fs_reg(brw_vec8_grf(first_icp_handle, 0)),
+ fs_reg(icp_offset_bytes),
+ brw_imm_ud(nir->info.gs.vertices_in * REG_SIZE));
+ }
+
+ fs_inst *inst;
+ if (offset_const) {
+ /* Constant indexing - use global offset. */
+ inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dst, icp_handle);
+ inst->offset = base_offset + offset_const->u32[0];
+ inst->base_mrf = -1;
+ inst->mlen = 1;
+ inst->regs_written = num_components;
} else {
- fs_inst *inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dst,
- fs_reg(vertex), fs_reg(input_offset));
+ /* Indirect indexing - use per-slot offsets as well. */
+ const fs_reg srcs[] = { icp_handle, get_nir_src(offset_src) };
+ fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, 2);
+ bld.LOAD_PAYLOAD(payload, srcs, ARRAY_SIZE(srcs), 0);
+
+ inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, dst, payload);
+ inst->offset = base_offset;
+ inst->base_mrf = -1;
+ inst->mlen = 2;
inst->regs_written = num_components;
}
+
+ if (is_point_size) {
+ /* Read the whole VUE header (because of alignment) and read .w. */
+ fs_reg tmp = bld.vgrf(dst.type, 4);
+ inst->dst = tmp;
+ inst->regs_written = 4;
+ bld.MOV(dst, offset(tmp, bld, 3));
+ }
+ }
+}
+
+fs_reg
+fs_visitor::get_indirect_offset(nir_intrinsic_instr *instr)
+{
+ nir_src *offset_src = nir_get_io_offset_src(instr);
+ nir_const_value *const_value = nir_src_as_const_value(*offset_src);
+
+ if (const_value) {
+ /* The only constant offset we should find is 0. brw_nir.c's
+ * add_const_offset_to_base() will fold other constant offsets
+ * into instr->const_index[0].
+ */
+ assert(const_value->u32[0] == 0);
+ return fs_reg();
}
+
+ return get_nir_src(*offset_src);
}
void
case nir_intrinsic_load_vertex_id_zero_base:
case nir_intrinsic_load_base_vertex:
- case nir_intrinsic_load_instance_id: {
+ case nir_intrinsic_load_instance_id:
+ case nir_intrinsic_load_base_instance:
+ case nir_intrinsic_load_draw_id: {
gl_system_value sv = nir_system_value_from_intrinsic(instr->intrinsic);
fs_reg val = nir_system_values[sv];
assert(val.file != BAD_FILE);
}
}
+void
+fs_visitor::nir_emit_tes_intrinsic(const fs_builder &bld,
+ nir_intrinsic_instr *instr)
+{
+ assert(stage == MESA_SHADER_TESS_EVAL);
+ struct brw_tes_prog_data *tes_prog_data = (struct brw_tes_prog_data *) prog_data;
+
+ fs_reg dest;
+ if (nir_intrinsic_infos[instr->intrinsic].has_dest)
+ dest = get_nir_dest(instr->dest);
+
+ switch (instr->intrinsic) {
+ case nir_intrinsic_load_primitive_id:
+ bld.MOV(dest, fs_reg(brw_vec1_grf(0, 1)));
+ break;
+ case nir_intrinsic_load_tess_coord:
+ /* gl_TessCoord is part of the payload in g1-3 */
+ for (unsigned i = 0; i < 3; i++) {
+ bld.MOV(offset(dest, bld, i), fs_reg(brw_vec8_grf(1 + i, 0)));
+ }
+ break;
+
+ case nir_intrinsic_load_tess_level_outer:
+ /* When the TES reads gl_TessLevelOuter, we ensure that the patch header
+ * appears as a push-model input. So, we can simply use the ATTR file
+ * rather than issuing URB read messages. The data is stored in the
+ * high DWords in reverse order - DWord 7 contains .x, DWord 6 contains
+ * .y, and so on.
+ */
+ switch (tes_prog_data->domain) {
+ case BRW_TESS_DOMAIN_QUAD:
+ for (unsigned i = 0; i < 4; i++)
+ bld.MOV(offset(dest, bld, i), component(fs_reg(ATTR, 0), 7 - i));
+ break;
+ case BRW_TESS_DOMAIN_TRI:
+ for (unsigned i = 0; i < 3; i++)
+ bld.MOV(offset(dest, bld, i), component(fs_reg(ATTR, 0), 7 - i));
+ break;
+ case BRW_TESS_DOMAIN_ISOLINE:
+ for (unsigned i = 0; i < 2; i++)
+ bld.MOV(offset(dest, bld, i), component(fs_reg(ATTR, 0), 7 - i));
+ break;
+ }
+ break;
+
+ case nir_intrinsic_load_tess_level_inner:
+ /* When the TES reads gl_TessLevelInner, we ensure that the patch header
+ * appears as a push-model input. So, we can simply use the ATTR file
+ * rather than issuing URB read messages.
+ */
+ switch (tes_prog_data->domain) {
+ case BRW_TESS_DOMAIN_QUAD:
+ bld.MOV(dest, component(fs_reg(ATTR, 0), 3));
+ bld.MOV(offset(dest, bld, 1), component(fs_reg(ATTR, 0), 2));
+ break;
+ case BRW_TESS_DOMAIN_TRI:
+ bld.MOV(dest, component(fs_reg(ATTR, 0), 4));
+ break;
+ case BRW_TESS_DOMAIN_ISOLINE:
+ /* ignore - value is undefined */
+ break;
+ }
+ break;
+
+ case nir_intrinsic_load_input:
+ case nir_intrinsic_load_per_vertex_input: {
+ fs_reg indirect_offset = get_indirect_offset(instr);
+ unsigned imm_offset = instr->const_index[0];
+
+ fs_inst *inst;
+ if (indirect_offset.file == BAD_FILE) {
+ /* Arbitrarily only push up to 32 vec4 slots worth of data,
+ * which is 16 registers (since each holds 2 vec4 slots).
+ */
+ const unsigned max_push_slots = 32;
+ if (imm_offset < max_push_slots) {
+ fs_reg src = fs_reg(ATTR, imm_offset / 2, dest.type);
+ for (int i = 0; i < instr->num_components; i++) {
+ bld.MOV(offset(dest, bld, i),
+ component(src, 4 * (imm_offset % 2) + i));
+ }
+ tes_prog_data->base.urb_read_length =
+ MAX2(tes_prog_data->base.urb_read_length,
+ DIV_ROUND_UP(imm_offset + 1, 2));
+ } else {
+ /* Replicate the patch handle to all enabled channels */
+ const fs_reg srcs[] = {
+ retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD)
+ };
+ fs_reg patch_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
+ bld.LOAD_PAYLOAD(patch_handle, srcs, ARRAY_SIZE(srcs), 0);
+
+ inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dest, patch_handle);
+ inst->mlen = 1;
+ inst->offset = imm_offset;
+ inst->base_mrf = -1;
+ inst->regs_written = instr->num_components;
+ }
+ } else {
+ /* Indirect indexing - use per-slot offsets as well. */
+ const fs_reg srcs[] = {
+ retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD),
+ indirect_offset
+ };
+ fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, 2);
+ bld.LOAD_PAYLOAD(payload, srcs, ARRAY_SIZE(srcs), 0);
+
+ inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, dest, payload);
+ inst->mlen = 2;
+ inst->offset = imm_offset;
+ inst->base_mrf = -1;
+ inst->regs_written = instr->num_components;
+ }
+ break;
+ }
+ default:
+ nir_emit_intrinsic(bld, instr);
+ break;
+ }
+}
+
void
fs_visitor::nir_emit_gs_intrinsic(const fs_builder &bld,
nir_intrinsic_instr *instr)
{
assert(stage == MESA_SHADER_GEOMETRY);
+ fs_reg indirect_offset;
fs_reg dest;
if (nir_intrinsic_infos[instr->intrinsic].has_dest)
retype(fs_reg(brw_vec8_grf(2, 0)), BRW_REGISTER_TYPE_UD));
break;
- case nir_intrinsic_load_input_indirect:
case nir_intrinsic_load_input:
unreachable("load_input intrinsics are invalid for the GS stage");
- case nir_intrinsic_load_per_vertex_input_indirect:
- assert(!"Not allowed");
case nir_intrinsic_load_per_vertex_input:
emit_gs_input_load(dest, instr->src[0], instr->const_index[0],
- instr->num_components);
+ instr->src[1], instr->num_components);
break;
case nir_intrinsic_emit_vertex_with_counter:
break;
}
+ case nir_intrinsic_load_helper_invocation:
case nir_intrinsic_load_sample_mask_in:
case nir_intrinsic_load_sample_id: {
gl_system_value sv = nir_system_value_from_intrinsic(instr->intrinsic);
fs_inst *cmp;
if (instr->intrinsic == nir_intrinsic_discard_if) {
cmp = bld.CMP(bld.null_reg_f(), get_nir_src(instr->src[0]),
- fs_reg(0), BRW_CONDITIONAL_Z);
+ brw_imm_d(0), BRW_CONDITIONAL_Z);
} else {
fs_reg some_reg = fs_reg(retype(brw_vec8_grf(0, 0),
BRW_REGISTER_TYPE_UW));
FS_OPCODE_INTERPOLATE_AT_CENTROID,
dst_xy,
fs_reg(), /* src */
- fs_reg(0u),
+ brw_imm_ud(0u),
interpolation);
break;
nir_const_value *const_sample = nir_src_as_const_value(instr->src[0]);
if (const_sample) {
- unsigned msg_data = const_sample->i[0] << 4;
+ unsigned msg_data = const_sample->i32[0] << 4;
emit_pixel_interpolater_send(bld,
FS_OPCODE_INTERPOLATE_AT_SAMPLE,
dst_xy,
fs_reg(), /* src */
- fs_reg(msg_data),
+ brw_imm_ud(msg_data),
interpolation);
} else {
const fs_reg sample_src = retype(get_nir_src(instr->src[0]),
if (nir_src_is_dynamically_uniform(instr->src[0])) {
const fs_reg sample_id = bld.emit_uniformize(sample_src);
const fs_reg msg_data = vgrf(glsl_type::uint_type);
- bld.exec_all().group(1, 0).SHL(msg_data, sample_id, fs_reg(4u));
+ bld.exec_all().group(1, 0)
+ .SHL(msg_data, sample_id, brw_imm_ud(4u));
emit_pixel_interpolater_send(bld,
FS_OPCODE_INTERPOLATE_AT_SAMPLE,
dst_xy,
sample_src, sample_id,
BRW_CONDITIONAL_EQ);
const fs_reg msg_data = vgrf(glsl_type::uint_type);
- bld.exec_all().group(1, 0).SHL(msg_data, sample_id, fs_reg(4u));
+ bld.exec_all().group(1, 0)
+ .SHL(msg_data, sample_id, brw_imm_ud(4u));
fs_inst *inst =
emit_pixel_interpolater_send(bld,
FS_OPCODE_INTERPOLATE_AT_SAMPLE,
nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
if (const_offset) {
- unsigned off_x = MIN2((int)(const_offset->f[0] * 16), 7) & 0xf;
- unsigned off_y = MIN2((int)(const_offset->f[1] * 16), 7) & 0xf;
+ unsigned off_x = MIN2((int)(const_offset->f32[0] * 16), 7) & 0xf;
+ unsigned off_y = MIN2((int)(const_offset->f32[1] * 16), 7) & 0xf;
emit_pixel_interpolater_send(bld,
FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET,
dst_xy,
fs_reg(), /* src */
- fs_reg(off_x | (off_y << 4)),
+ brw_imm_ud(off_x | (off_y << 4)),
interpolation);
} else {
fs_reg src = vgrf(glsl_type::ivec2_type);
BRW_REGISTER_TYPE_F);
for (int i = 0; i < 2; i++) {
fs_reg temp = vgrf(glsl_type::float_type);
- bld.MUL(temp, offset(offset_src, bld, i), fs_reg(16.0f));
+ bld.MUL(temp, offset(offset_src, bld, i), brw_imm_f(16.0f));
fs_reg itemp = vgrf(glsl_type::int_type);
bld.MOV(itemp, temp); /* float to int */
* FRAGMENT_INTERPOLATION_OFFSET_BITS"
*/
set_condmod(BRW_CONDITIONAL_L,
- bld.SEL(offset(src, bld, i), itemp, fs_reg(7)));
+ bld.SEL(offset(src, bld, i), itemp, brw_imm_d(7)));
}
const enum opcode opcode = FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET;
opcode,
dst_xy,
src,
- fs_reg(0u),
+ brw_imm_ud(0u),
interpolation);
}
break;
cs_prog_data->uses_num_work_groups = true;
- fs_reg surf_index = fs_reg(surface);
+ fs_reg surf_index = brw_imm_ud(surface);
brw_mark_surface_used(prog_data, surface);
/* Read the 3 GLuint components of gl_NumWorkGroups */
for (unsigned i = 0; i < 3; i++) {
fs_reg read_result =
emit_untyped_read(bld, surf_index,
- fs_reg(i << 2),
+ brw_imm_ud(i << 2),
1 /* dims */, 1 /* size */,
BRW_PREDICATE_NONE);
read_result.type = dest.type;
break;
}
+ case nir_intrinsic_shared_atomic_add:
+ nir_emit_shared_atomic(bld, BRW_AOP_ADD, instr);
+ break;
+ case nir_intrinsic_shared_atomic_imin:
+ nir_emit_shared_atomic(bld, BRW_AOP_IMIN, instr);
+ break;
+ case nir_intrinsic_shared_atomic_umin:
+ nir_emit_shared_atomic(bld, BRW_AOP_UMIN, instr);
+ break;
+ case nir_intrinsic_shared_atomic_imax:
+ nir_emit_shared_atomic(bld, BRW_AOP_IMAX, instr);
+ break;
+ case nir_intrinsic_shared_atomic_umax:
+ nir_emit_shared_atomic(bld, BRW_AOP_UMAX, instr);
+ break;
+ case nir_intrinsic_shared_atomic_and:
+ nir_emit_shared_atomic(bld, BRW_AOP_AND, instr);
+ break;
+ case nir_intrinsic_shared_atomic_or:
+ nir_emit_shared_atomic(bld, BRW_AOP_OR, instr);
+ break;
+ case nir_intrinsic_shared_atomic_xor:
+ nir_emit_shared_atomic(bld, BRW_AOP_XOR, instr);
+ break;
+ case nir_intrinsic_shared_atomic_exchange:
+ nir_emit_shared_atomic(bld, BRW_AOP_MOV, instr);
+ break;
+ case nir_intrinsic_shared_atomic_comp_swap:
+ nir_emit_shared_atomic(bld, BRW_AOP_CMPWR, instr);
+ break;
+
+ case nir_intrinsic_load_shared: {
+ assert(devinfo->gen >= 7);
+
+ fs_reg surf_index = brw_imm_ud(GEN7_BTI_SLM);
+
+ /* Get the offset to read from */
+ fs_reg offset_reg;
+ nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
+ if (const_offset) {
+ offset_reg = brw_imm_ud(instr->const_index[0] + const_offset->u32[0]);
+ } else {
+ offset_reg = vgrf(glsl_type::uint_type);
+ bld.ADD(offset_reg,
+ retype(get_nir_src(instr->src[0]), BRW_REGISTER_TYPE_UD),
+ brw_imm_ud(instr->const_index[0]));
+ }
+
+ /* Read the vector */
+ fs_reg read_result = emit_untyped_read(bld, surf_index, offset_reg,
+ 1 /* dims */,
+ instr->num_components,
+ BRW_PREDICATE_NONE);
+ read_result.type = dest.type;
+ for (int i = 0; i < instr->num_components; i++)
+ bld.MOV(offset(dest, bld, i), offset(read_result, bld, i));
+
+ break;
+ }
+
+ case nir_intrinsic_store_shared: {
+ assert(devinfo->gen >= 7);
+
+ /* Block index */
+ fs_reg surf_index = brw_imm_ud(GEN7_BTI_SLM);
+
+ /* Value */
+ fs_reg val_reg = get_nir_src(instr->src[0]);
+
+ /* Writemask */
+ unsigned writemask = instr->const_index[1];
+
+ /* Combine groups of consecutive enabled channels in one write
+ * message. We use ffs to find the first enabled channel and then ffs on
+ * the bit-inverse, down-shifted writemask to determine the length of
+ * the block of enabled bits.
+ */
+ while (writemask) {
+ unsigned first_component = ffs(writemask) - 1;
+ unsigned length = ffs(~(writemask >> first_component)) - 1;
+ fs_reg offset_reg;
+
+ nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
+ if (const_offset) {
+ offset_reg = brw_imm_ud(instr->const_index[0] + const_offset->u32[0] +
+ 4 * first_component);
+ } else {
+ offset_reg = vgrf(glsl_type::uint_type);
+ bld.ADD(offset_reg,
+ retype(get_nir_src(instr->src[1]), BRW_REGISTER_TYPE_UD),
+ brw_imm_ud(instr->const_index[0] + 4 * first_component));
+ }
+
+ emit_untyped_write(bld, surf_index, offset_reg,
+ offset(val_reg, bld, first_component),
+ 1 /* dims */, length,
+ BRW_PREDICATE_NONE);
+
+ /* Clear the bits in the writemask that we just wrote, then try
+ * again to see if more channels are left.
+ */
+ writemask &= (15 << (first_component + length));
+ }
+
+ break;
+ }
+
default:
nir_emit_intrinsic(bld, instr);
break;
if (nir_intrinsic_infos[instr->intrinsic].has_dest)
dest = get_nir_dest(instr->dest);
- bool has_indirect = false;
-
switch (instr->intrinsic) {
case nir_intrinsic_atomic_counter_inc:
case nir_intrinsic_atomic_counter_dec:
case nir_intrinsic_atomic_counter_read: {
- using namespace surface_access;
-
/* Get the arguments of the atomic intrinsic. */
const fs_reg offset = get_nir_src(instr->src[0]);
const unsigned surface = (stage_prog_data->binding_table.abo_start +
/* Emit a surface read or atomic op. */
switch (instr->intrinsic) {
case nir_intrinsic_atomic_counter_read:
- tmp = emit_untyped_read(bld, fs_reg(surface), offset, 1, 1);
+ tmp = emit_untyped_read(bld, brw_imm_ud(surface), offset, 1, 1);
break;
case nir_intrinsic_atomic_counter_inc:
- tmp = emit_untyped_atomic(bld, fs_reg(surface), offset, fs_reg(),
+ tmp = emit_untyped_atomic(bld, brw_imm_ud(surface), offset, fs_reg(),
fs_reg(), 1, 1, BRW_AOP_INC);
break;
case nir_intrinsic_atomic_counter_dec:
- tmp = emit_untyped_atomic(bld, fs_reg(surface), offset, fs_reg(),
+ tmp = emit_untyped_atomic(bld, brw_imm_ud(surface), offset, fs_reg(),
fs_reg(), 1, 1, BRW_AOP_PREDEC);
break;
for (unsigned c = 0; c < info->dest_components; ++c) {
if ((int)c >= type->coordinate_components()) {
bld.MOV(offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
- fs_reg(1));
+ brw_imm_d(1));
} else if (c == 1 && is_1d_array_image) {
bld.MOV(offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
offset(size, bld, 2));
} else if (c == 2 && is_cube_array_image) {
bld.emit(SHADER_OPCODE_INT_QUOTIENT,
offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
- offset(size, bld, c), fs_reg(6));
+ offset(size, bld, c), brw_imm_d(6));
} else {
bld.MOV(offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
offset(size, bld, c));
case nir_intrinsic_image_samples:
/* The driver does not support multi-sampled images. */
- bld.MOV(retype(dest, BRW_REGISTER_TYPE_D), fs_reg(1));
+ bld.MOV(retype(dest, BRW_REGISTER_TYPE_D), brw_imm_d(1));
break;
- case nir_intrinsic_load_uniform_indirect:
- has_indirect = true;
- /* fallthrough */
case nir_intrinsic_load_uniform: {
- fs_reg uniform_reg(UNIFORM, instr->const_index[0]);
- uniform_reg.reg_offset = instr->const_index[1];
+ /* Offsets are in bytes but they should always be multiples of 4 */
+ assert(instr->const_index[0] % 4 == 0);
- for (unsigned j = 0; j < instr->num_components; j++) {
- fs_reg src = offset(retype(uniform_reg, dest.type), bld, j);
- if (has_indirect)
- src.reladdr = new(mem_ctx) fs_reg(get_nir_src(instr->src[0]));
+ fs_reg src(UNIFORM, instr->const_index[0] / 4, dest.type);
- bld.MOV(dest, src);
- dest = offset(dest, bld, 1);
+ nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
+ if (const_offset) {
+ /* Offsets are in bytes but they should always be multiples of 4 */
+ assert(const_offset->u32[0] % 4 == 0);
+ src.reg_offset = const_offset->u32[0] / 4;
+
+ for (unsigned j = 0; j < instr->num_components; j++) {
+ bld.MOV(offset(dest, bld, j), offset(src, bld, j));
+ }
+ } else {
+ fs_reg indirect = retype(get_nir_src(instr->src[0]),
+ BRW_REGISTER_TYPE_UD);
+
+ /* We need to pass a size to the MOV_INDIRECT but we don't want it to
+ * go past the end of the uniform. In order to keep the n'th
+ * component from running past, we subtract off the size of all but
+ * one component of the vector.
+ */
+ assert(instr->const_index[1] >= instr->num_components * 4);
+ unsigned read_size = instr->const_index[1] -
+ (instr->num_components - 1) * 4;
+
+ for (unsigned j = 0; j < instr->num_components; j++) {
+ bld.emit(SHADER_OPCODE_MOV_INDIRECT,
+ offset(dest, bld, j), offset(src, bld, j),
+ indirect, brw_imm_ud(read_size));
+ }
}
break;
}
- case nir_intrinsic_load_ubo_indirect:
- has_indirect = true;
- /* fallthrough */
case nir_intrinsic_load_ubo: {
nir_const_value *const_index = nir_src_as_const_value(instr->src[0]);
fs_reg surf_index;
if (const_index) {
const unsigned index = stage_prog_data->binding_table.ubo_start +
- const_index->u[0];
- surf_index = fs_reg(index);
+ const_index->u32[0];
+ surf_index = brw_imm_ud(index);
brw_mark_surface_used(prog_data, index);
} else {
/* The block index is not a constant. Evaluate the index expression
*/
surf_index = vgrf(glsl_type::uint_type);
bld.ADD(surf_index, get_nir_src(instr->src[0]),
- fs_reg(stage_prog_data->binding_table.ubo_start));
+ brw_imm_ud(stage_prog_data->binding_table.ubo_start));
surf_index = bld.emit_uniformize(surf_index);
/* Assume this may touch any UBO. It would be nice to provide
nir->info.num_ubos - 1);
}
- if (has_indirect) {
- /* Turn the byte offset into a dword offset. */
- fs_reg base_offset = vgrf(glsl_type::int_type);
- bld.SHR(base_offset, retype(get_nir_src(instr->src[1]),
- BRW_REGISTER_TYPE_D),
- fs_reg(2));
+ nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
+ if (const_offset == NULL) {
+ fs_reg base_offset = retype(get_nir_src(instr->src[1]),
+ BRW_REGISTER_TYPE_D);
- unsigned vec4_offset = instr->const_index[0] / 4;
for (int i = 0; i < instr->num_components; i++)
VARYING_PULL_CONSTANT_LOAD(bld, offset(dest, bld, i), surf_index,
- base_offset, vec4_offset + i);
+ base_offset, i * 4);
} else {
fs_reg packed_consts = vgrf(glsl_type::float_type);
packed_consts.type = dest.type;
- fs_reg const_offset_reg((unsigned) instr->const_index[0] & ~15);
+ struct brw_reg const_offset_reg = brw_imm_ud(const_offset->u32[0] & ~15);
bld.emit(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD, packed_consts,
surf_index, const_offset_reg);
for (unsigned i = 0; i < instr->num_components; i++) {
- packed_consts.set_smear(instr->const_index[0] % 16 / 4 + i);
+ packed_consts.set_smear(const_offset->u32[0] % 16 / 4 + i);
/* The std140 packing rules don't allow vectors to cross 16-byte
* boundaries, and a reg is 32 bytes.
break;
}
- case nir_intrinsic_load_ssbo_indirect:
- has_indirect = true;
- /* fallthrough */
case nir_intrinsic_load_ssbo: {
assert(devinfo->gen >= 7);
fs_reg surf_index;
if (const_uniform_block) {
unsigned index = stage_prog_data->binding_table.ssbo_start +
- const_uniform_block->u[0];
- surf_index = fs_reg(index);
+ const_uniform_block->u32[0];
+ surf_index = brw_imm_ud(index);
brw_mark_surface_used(prog_data, index);
} else {
surf_index = vgrf(glsl_type::uint_type);
bld.ADD(surf_index, get_nir_src(instr->src[0]),
- fs_reg(stage_prog_data->binding_table.ssbo_start));
+ brw_imm_ud(stage_prog_data->binding_table.ssbo_start));
/* Assume this may touch any UBO. It would be nice to provide
* a tighter bound, but the array information is already lowered away.
nir->info.num_ssbos - 1);
}
- /* Get the offset to read from */
fs_reg offset_reg;
- if (has_indirect) {
- offset_reg = get_nir_src(instr->src[1]);
+ nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
+ if (const_offset) {
+ offset_reg = brw_imm_ud(const_offset->u32[0]);
} else {
- offset_reg = fs_reg(instr->const_index[0]);
+ offset_reg = get_nir_src(instr->src[1]);
}
/* Read the vector */
break;
}
- case nir_intrinsic_load_input_indirect:
- has_indirect = true;
- /* fallthrough */
case nir_intrinsic_load_input: {
- unsigned index = 0;
- for (unsigned j = 0; j < instr->num_components; j++) {
- fs_reg src;
- if (stage == MESA_SHADER_VERTEX) {
- src = offset(fs_reg(ATTR, instr->const_index[0], dest.type), bld, index);
- } else {
- src = offset(retype(nir_inputs, dest.type), bld,
- instr->const_index[0] + index);
- }
- if (has_indirect)
- src.reladdr = new(mem_ctx) fs_reg(get_nir_src(instr->src[0]));
- index++;
+ fs_reg src;
+ if (stage == MESA_SHADER_VERTEX) {
+ src = fs_reg(ATTR, instr->const_index[0], dest.type);
+ } else {
+ src = offset(retype(nir_inputs, dest.type), bld,
+ instr->const_index[0]);
+ }
- bld.MOV(dest, src);
- dest = offset(dest, bld, 1);
+ nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
+ assert(const_offset && "Indirect input loads not allowed");
+ src = offset(src, bld, const_offset->u32[0]);
+
+ for (unsigned j = 0; j < instr->num_components; j++) {
+ bld.MOV(offset(dest, bld, j), offset(src, bld, j));
}
break;
}
- case nir_intrinsic_store_ssbo_indirect:
- has_indirect = true;
- /* fallthrough */
case nir_intrinsic_store_ssbo: {
assert(devinfo->gen >= 7);
nir_src_as_const_value(instr->src[1]);
if (const_uniform_block) {
unsigned index = stage_prog_data->binding_table.ssbo_start +
- const_uniform_block->u[0];
- surf_index = fs_reg(index);
+ const_uniform_block->u32[0];
+ surf_index = brw_imm_ud(index);
brw_mark_surface_used(prog_data, index);
} else {
surf_index = vgrf(glsl_type::uint_type);
bld.ADD(surf_index, get_nir_src(instr->src[1]),
- fs_reg(stage_prog_data->binding_table.ssbo_start));
+ brw_imm_ud(stage_prog_data->binding_table.ssbo_start));
brw_mark_surface_used(prog_data,
stage_prog_data->binding_table.ssbo_start +
fs_reg val_reg = get_nir_src(instr->src[0]);
/* Writemask */
- unsigned writemask = instr->const_index[1];
+ unsigned writemask = instr->const_index[0];
/* Combine groups of consecutive enabled channels in one write
* message. We use ffs to find the first enabled channel and then ffs on
while (writemask) {
unsigned first_component = ffs(writemask) - 1;
unsigned length = ffs(~(writemask >> first_component)) - 1;
- fs_reg offset_reg;
- if (!has_indirect) {
- offset_reg = fs_reg(instr->const_index[0] + 4 * first_component);
+ fs_reg offset_reg;
+ nir_const_value *const_offset = nir_src_as_const_value(instr->src[2]);
+ if (const_offset) {
+ offset_reg = brw_imm_ud(const_offset->u32[0] + 4 * first_component);
} else {
offset_reg = vgrf(glsl_type::uint_type);
bld.ADD(offset_reg,
retype(get_nir_src(instr->src[2]), BRW_REGISTER_TYPE_UD),
- fs_reg(4 * first_component));
+ brw_imm_ud(4 * first_component));
}
emit_untyped_write(bld, surf_index, offset_reg,
break;
}
- case nir_intrinsic_store_output_indirect:
- has_indirect = true;
- /* fallthrough */
case nir_intrinsic_store_output: {
fs_reg src = get_nir_src(instr->src[0]);
- unsigned index = 0;
+ fs_reg new_dest = offset(retype(nir_outputs, src.type), bld,
+ instr->const_index[0]);
+
+ nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
+ assert(const_offset && "Indirect output stores not allowed");
+ new_dest = offset(new_dest, bld, const_offset->u32[0]);
+
for (unsigned j = 0; j < instr->num_components; j++) {
- fs_reg new_dest = offset(retype(nir_outputs, src.type), bld,
- instr->const_index[0] + index);
- if (has_indirect)
- src.reladdr = new(mem_ctx) fs_reg(get_nir_src(instr->src[1]));
- index++;
- bld.MOV(new_dest, src);
- src = offset(src, bld, 1);
+ bld.MOV(offset(new_dest, bld, j), offset(src, bld, j));
}
break;
}
case nir_intrinsic_get_buffer_size: {
nir_const_value *const_uniform_block = nir_src_as_const_value(instr->src[0]);
- unsigned ssbo_index = const_uniform_block ? const_uniform_block->u[0] : 0;
+ unsigned ssbo_index = const_uniform_block ? const_uniform_block->u32[0] : 0;
int reg_width = dispatch_width / 8;
/* Set LOD = 0 */
- fs_reg source = fs_reg(0);
+ fs_reg source = brw_imm_d(0);
int mlen = 1 * reg_width;
* SURFTYPE_BUFFER.
*/
int regs_written = 4 * mlen;
- fs_reg src_payload = fs_reg(GRF, alloc.allocate(mlen),
+ fs_reg src_payload = fs_reg(VGRF, alloc.allocate(mlen),
BRW_REGISTER_TYPE_UD);
bld.LOAD_PAYLOAD(src_payload, &source, 1, 0);
- fs_reg buffer_size = fs_reg(GRF, alloc.allocate(regs_written),
+ fs_reg buffer_size = fs_reg(VGRF, alloc.allocate(regs_written),
BRW_REGISTER_TYPE_UD);
const unsigned index = prog_data->binding_table.ssbo_start + ssbo_index;
fs_inst *inst = bld.emit(FS_OPCODE_GET_BUFFER_SIZE, buffer_size,
- src_payload, fs_reg(index));
+ src_payload, brw_imm_ud(index));
inst->header_size = 0;
inst->mlen = mlen;
inst->regs_written = regs_written;
nir_const_value *const_surface = nir_src_as_const_value(instr->src[0]);
if (const_surface) {
unsigned surf_index = stage_prog_data->binding_table.ssbo_start +
- const_surface->u[0];
- surface = fs_reg(surf_index);
+ const_surface->u32[0];
+ surface = brw_imm_ud(surf_index);
brw_mark_surface_used(prog_data, surf_index);
} else {
surface = vgrf(glsl_type::uint_type);
bld.ADD(surface, get_nir_src(instr->src[0]),
- fs_reg(stage_prog_data->binding_table.ssbo_start));
+ brw_imm_ud(stage_prog_data->binding_table.ssbo_start));
/* Assume this may touch any SSBO. This is the same we do for other
* UBO/SSBO accesses with non-constant surface.
/* Emit the actual atomic operation operation */
- fs_reg atomic_result =
- surface_access::emit_untyped_atomic(bld, surface, offset,
- data1, data2,
- 1 /* dims */, 1 /* rsize */,
- op,
- BRW_PREDICATE_NONE);
+ fs_reg atomic_result = emit_untyped_atomic(bld, surface, offset,
+ data1, data2,
+ 1 /* dims */, 1 /* rsize */,
+ op,
+ BRW_PREDICATE_NONE);
+ dest.type = atomic_result.type;
+ bld.MOV(dest, atomic_result);
+}
+
+void
+fs_visitor::nir_emit_shared_atomic(const fs_builder &bld,
+ int op, nir_intrinsic_instr *instr)
+{
+ fs_reg dest;
+ if (nir_intrinsic_infos[instr->intrinsic].has_dest)
+ dest = get_nir_dest(instr->dest);
+
+ fs_reg surface = brw_imm_ud(GEN7_BTI_SLM);
+ fs_reg offset = get_nir_src(instr->src[0]);
+ fs_reg data1 = get_nir_src(instr->src[1]);
+ fs_reg data2;
+ if (op == BRW_AOP_CMPWR)
+ data2 = get_nir_src(instr->src[2]);
+
+ /* Emit the actual atomic operation operation */
+
+ fs_reg atomic_result = emit_untyped_atomic(bld, surface, offset,
+ data1, data2,
+ 1 /* dims */, 1 /* rsize */,
+ op,
+ BRW_PREDICATE_NONE);
dest.type = atomic_result.type;
bld.MOV(dest, atomic_result);
}
void
fs_visitor::nir_emit_texture(const fs_builder &bld, nir_tex_instr *instr)
{
+ unsigned texture = instr->texture_index;
unsigned sampler = instr->sampler_index;
- fs_reg sampler_reg(sampler);
+ fs_reg texture_reg(brw_imm_ud(texture));
+ fs_reg sampler_reg(brw_imm_ud(sampler));
int gather_component = instr->component;
- bool is_rect = instr->sampler_dim == GLSL_SAMPLER_DIM_RECT;
-
bool is_cube_array = instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE &&
instr->is_array;
int lod_components = 0;
- int UNUSED offset_components = 0;
fs_reg coordinate, shadow_comparitor, lod, lod2, sample_index, mcs, tex_offset;
+ /* The hardware requires a LOD for buffer textures */
+ if (instr->sampler_dim == GLSL_SAMPLER_DIM_BUF)
+ lod = brw_imm_d(0);
+
for (unsigned i = 0; i < instr->num_srcs; i++) {
fs_reg src = get_nir_src(instr->src[i].src);
switch (instr->src[i].src_type) {
switch (instr->op) {
case nir_texop_txf:
case nir_texop_txf_ms:
+ case nir_texop_samples_identical:
coordinate = retype(src, BRW_REGISTER_TYPE_D);
break;
default:
case nir_tex_src_ms_index:
sample_index = retype(src, BRW_REGISTER_TYPE_UD);
break;
- case nir_tex_src_offset:
- tex_offset = retype(src, BRW_REGISTER_TYPE_D);
- if (instr->is_array)
- offset_components = instr->coord_components - 1;
- else
- offset_components = instr->coord_components;
+
+ case nir_tex_src_offset: {
+ nir_const_value *const_offset =
+ nir_src_as_const_value(instr->src[i].src);
+ if (const_offset) {
+ tex_offset = brw_imm_ud(brw_texture_offset(const_offset->i32, 3));
+ } else {
+ tex_offset = retype(src, BRW_REGISTER_TYPE_D);
+ }
break;
+ }
+
case nir_tex_src_projector:
unreachable("should be lowered");
- case nir_tex_src_sampler_offset: {
- /* Figure out the highest possible sampler index and mark it as used */
- uint32_t max_used = sampler + instr->sampler_array_size - 1;
+ case nir_tex_src_texture_offset: {
+ /* Figure out the highest possible texture index and mark it as used */
+ uint32_t max_used = texture + instr->texture_array_size - 1;
if (instr->op == nir_texop_tg4 && devinfo->gen < 8) {
max_used += stage_prog_data->binding_table.gather_texture_start;
} else {
}
brw_mark_surface_used(prog_data, max_used);
+ /* Emit code to evaluate the actual indexing expression */
+ texture_reg = vgrf(glsl_type::uint_type);
+ bld.ADD(texture_reg, src, brw_imm_ud(texture));
+ texture_reg = bld.emit_uniformize(texture_reg);
+ break;
+ }
+
+ case nir_tex_src_sampler_offset: {
/* Emit code to evaluate the actual indexing expression */
sampler_reg = vgrf(glsl_type::uint_type);
- bld.ADD(sampler_reg, src, fs_reg(sampler));
+ bld.ADD(sampler_reg, src, brw_imm_ud(sampler));
sampler_reg = bld.emit_uniformize(sampler_reg);
break;
}
}
}
- if (instr->op == nir_texop_txf_ms) {
+ if (instr->op == nir_texop_txf_ms ||
+ instr->op == nir_texop_samples_identical) {
if (devinfo->gen >= 7 &&
- key_tex->compressed_multisample_layout_mask & (1 << sampler)) {
- mcs = emit_mcs_fetch(coordinate, instr->coord_components, sampler_reg);
+ key_tex->compressed_multisample_layout_mask & (1 << texture)) {
+ mcs = emit_mcs_fetch(coordinate, instr->coord_components, texture_reg);
} else {
- mcs = fs_reg(0u);
- }
- }
-
- for (unsigned i = 0; i < 3; i++) {
- if (instr->const_offset[i] != 0) {
- assert(offset_components == 0);
- tex_offset = fs_reg(brw_texture_offset(instr->const_offset, 3));
- break;
+ mcs = brw_imm_ud(0u);
}
}
fs_reg dst = retype(get_nir_dest(instr->dest), BRW_REGISTER_TYPE_D);
fs_inst *inst = bld.emit(SHADER_OPCODE_SAMPLEINFO, dst,
bld.vgrf(BRW_REGISTER_TYPE_D, 1),
- sampler_reg);
+ texture_reg, texture_reg);
inst->mlen = 1;
inst->header_size = 1;
inst->base_mrf = -1;
return;
}
+ case nir_texop_samples_identical: op = ir_samples_identical; break;
default:
unreachable("unknown texture opcode");
}
emit_texture(op, dest_type, coordinate, instr->coord_components,
shadow_comparitor, lod, lod2, lod_components, sample_index,
- tex_offset, mcs, gather_component,
- is_cube_array, is_rect, sampler, sampler_reg);
+ tex_offset, mcs, gather_component, is_cube_array,
+ texture, texture_reg, sampler, sampler_reg);
fs_reg dest = get_nir_dest(instr->dest);
dest.type = this->result.type;