#include "compiler/glsl/ir.h"
#include "brw_fs.h"
-#include "brw_fs_surface_builder.h"
#include "brw_nir.h"
+#include "util/u_math.h"
+#include "util/bitscan.h"
using namespace brw;
-using namespace brw::surface_access;
void
fs_visitor::emit_nir_code()
nir_setup_uniforms();
nir_emit_system_values();
- /* get the main function and emit it */
- nir_foreach_function(function, nir) {
- assert(strcmp(function->name, "main") == 0);
- assert(function->impl);
- nir_emit_impl(function->impl);
- }
+ nir_emit_impl(nir_shader_get_entrypoint((nir_shader *)nir));
}
void
vec4s[loc] = MAX2(vec4s[loc], var_vec4s);
}
- nir_foreach_variable(var, &nir->outputs) {
- const int loc = var->data.driver_location;
- if (outputs[loc].file == BAD_FILE) {
- fs_reg reg = bld.vgrf(BRW_REGISTER_TYPE_F, 4 * vec4s[loc]);
- for (unsigned i = 0; i < vec4s[loc]; i++) {
- outputs[loc + i] = offset(reg, bld, 4 * i);
- }
+ for (unsigned loc = 0; loc < ARRAY_SIZE(vec4s);) {
+ if (vec4s[loc] == 0) {
+ loc++;
+ continue;
}
+
+ unsigned reg_size = vec4s[loc];
+
+ /* Check if there are any ranges that start within this range and extend
+ * past it. If so, include them in this allocation.
+ */
+ for (unsigned i = 1; i < reg_size; i++)
+ reg_size = MAX2(vec4s[i + loc] + i, reg_size);
+
+ fs_reg reg = bld.vgrf(BRW_REGISTER_TYPE_F, 4 * reg_size);
+ for (unsigned i = 0; i < reg_size; i++)
+ outputs[loc + i] = offset(reg, bld, 4 * i);
+
+ loc += reg_size;
}
}
* 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_v(0x76543210));
+
+ for (unsigned i = 0; i < DIV_ROUND_UP(v->dispatch_width, 16); i++) {
+ const fs_builder hbld = abld.group(MIN2(16, v->dispatch_width), i);
+ hbld.SHR(offset(shifted, hbld, i),
+ stride(retype(brw_vec1_grf(1 + i, 7),
+ BRW_REGISTER_TYPE_UB),
+ 1, 8, 0),
+ brw_imm_v(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
}
}
- nir_foreach_function(function, nir) {
- assert(strcmp(function->name, "main") == 0);
- assert(function->impl);
- nir_foreach_block(block, function->impl) {
- emit_system_values_block(block, this);
- }
- }
+ nir_function_impl *impl = nir_shader_get_entrypoint((nir_shader *)nir);
+ nir_foreach_block(block, impl)
+ emit_system_values_block(block, this);
}
/*
default:
unreachable("Invalid bit size");
}
+ case BRW_REGISTER_TYPE_B:
case BRW_REGISTER_TYPE_W:
case BRW_REGISTER_TYPE_D:
case BRW_REGISTER_TYPE_Q:
switch(bit_size) {
+ case 8:
+ return BRW_REGISTER_TYPE_B;
case 16:
return BRW_REGISTER_TYPE_W;
case 32:
default:
unreachable("Invalid bit size");
}
+ case BRW_REGISTER_TYPE_UB:
case BRW_REGISTER_TYPE_UW:
case BRW_REGISTER_TYPE_UD:
case BRW_REGISTER_TYPE_UQ:
switch(bit_size) {
+ case 8:
+ return BRW_REGISTER_TYPE_UB;
case 16:
return BRW_REGISTER_TYPE_UW;
case 32:
nir_emit_cf_list(&if_stmt->else_list);
bld.emit(BRW_OPCODE_ENDIF);
+
+ if (devinfo->gen < 7)
+ limit_dispatch_width(16, "Non-uniform control flow unsupported "
+ "in SIMD32 mode.");
}
void
nir_emit_cf_list(&loop->body);
bld.emit(BRW_OPCODE_WHILE);
+
+ if (devinfo->gen < 7)
+ limit_dispatch_width(16, "Non-uniform control flow unsupported "
+ "in SIMD32 mode.");
}
void
nir_emit_alu(abld, nir_instr_as_alu(instr));
break;
+ case nir_instr_type_deref:
+ /* Derefs can exist for images but they do nothing */
+ break;
+
case nir_instr_type_intrinsic:
switch (stage) {
case MESA_SHADER_VERTEX:
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);
+ /* If either opcode has source modifiers, bail.
+ *
+ * TODO: We can potentially handle source modifiers if both of the opcodes
+ * we're combining are signed integers.
+ */
+ if (instr->src[0].abs || instr->src[0].negate ||
+ src0->src[0].abs || src0->src[0].negate)
+ return false;
+
+ unsigned element = nir_src_as_uint(src0->src[1].src);
/* Element type to extract.*/
const brw_reg_type type = brw_int_type(
op0 = offset(op0, bld, src0->src[0].swizzle[0]);
set_saturate(instr->dest.saturate,
- bld.MOV(result, subscript(op0, type, element->u32[0])));
+ bld.MOV(result, subscript(op0, type, element)));
return true;
}
if (src0->intrinsic != nir_intrinsic_load_front_face)
return false;
- nir_const_value *value1 = nir_src_as_const_value(instr->src[1].src);
- if (!value1 || fabsf(value1->f32[0]) != 1.0f)
+ if (!nir_src_is_const(instr->src[1].src) ||
+ !nir_src_is_const(instr->src[2].src))
return false;
- nir_const_value *value2 = nir_src_as_const_value(instr->src[2].src);
- if (!value2 || fabsf(value2->f32[0]) != 1.0f)
+ const float value1 = nir_src_as_float(instr->src[1].src);
+ const float value2 = nir_src_as_float(instr->src[2].src);
+ if (fabsf(value1) != 1.0f || fabsf(value2) != 1.0f)
return false;
+ /* nir_opt_algebraic should have gotten rid of bcsel(b, a, a) */
+ assert(value1 == -value2);
+
fs_reg tmp = vgrf(glsl_type::int_type);
if (devinfo->gen >= 6) {
* surely be TRIANGLES
*/
- if (value1->f32[0] == -1.0f) {
+ if (value1 == -1.0f) {
g0.negate = true;
}
* surely be TRIANGLES
*/
- if (value1->f32[0] == -1.0f) {
+ if (value1 == -1.0f) {
g1_6.negate = true;
}
/* Since NIR is doing the scalarizing for us, we should only ever see
* vectorized operations with a single channel.
*/
- assert(_mesa_bitcount(instr->dest.write_mask) == 1);
+ assert(util_bitcount(instr->dest.write_mask) == 1);
channel = ffs(instr->dest.write_mask) - 1;
result = offset(result, bld, channel);
* BRW_OPCODE_F32TO16 when/if we work for HF support on gen7.
*/
- case nir_op_f2f16_undef:
+ case nir_op_f2f16:
inst = bld.MOV(result, op[0]);
inst->saturate = instr->dest.saturate;
break;
+ case nir_op_b2i8:
+ case nir_op_b2i16:
+ case nir_op_b2i32:
+ case nir_op_b2i64:
+ case nir_op_b2f16:
+ case nir_op_b2f32:
+ case nir_op_b2f64:
+ op[0].type = BRW_REGISTER_TYPE_D;
+ op[0].negate = !op[0].negate;
+ /* fallthrough */
case nir_op_f2f64:
case nir_op_f2i64:
case nir_op_f2u64:
case nir_op_i2i64:
case nir_op_u2f64:
case nir_op_u2u64:
- /* CHV PRM, vol07, 3D Media GPGPU Engine, Register Region Restrictions:
- *
- * "When source or destination is 64b (...), regioning in Align1
- * must follow these rules:
- *
- * 1. Source and destination horizontal stride must be aligned to
- * the same qword.
- * (...)"
- *
- * This means that conversions from bit-sizes smaller than 64-bit to
- * 64-bit need to have the source data elements aligned to 64-bit.
- * This restriction does not apply to BDW and later.
- */
- if (nir_dest_bit_size(instr->dest.dest) == 64 &&
- nir_src_bit_size(instr->src[0].src) < 64 &&
- (devinfo->is_cherryview || gen_device_info_is_9lp(devinfo))) {
- fs_reg tmp = bld.vgrf(result.type, 1);
- tmp = subscript(tmp, op[0].type, 0);
- inst = bld.MOV(tmp, op[0]);
- inst = bld.MOV(result, tmp);
- inst->saturate = instr->dest.saturate;
- break;
- }
- /* fallthrough */
case nir_op_f2f32:
case nir_op_f2i32:
case nir_op_f2u32:
case nir_op_u2u16:
case nir_op_i2f16:
case nir_op_u2f16:
+ case nir_op_i2i8:
+ case nir_op_u2u8:
inst = bld.MOV(result, op[0]);
inst->saturate = instr->dest.saturate;
break;
case nir_op_fsign: {
+ assert(!instr->dest.saturate);
if (op[0].abs) {
- /* Straightforward since the source can be assumed to be
- * non-negative.
+ /* Straightforward since the source can be assumed to be either
+ * strictly >= 0 or strictly <= 0 depending on the setting of the
+ * negate flag.
*/
set_condmod(BRW_CONDITIONAL_NZ, bld.MOV(result, op[0]));
- set_predicate(BRW_PREDICATE_NORMAL, bld.MOV(result, brw_imm_f(1.0f)));
+ inst = (op[0].negate)
+ ? bld.MOV(result, brw_imm_f(-1.0f))
+ : bld.MOV(result, brw_imm_f(1.0f));
+
+ set_predicate(BRW_PREDICATE_NORMAL, inst);
} else if (type_sz(op[0].type) < 8) {
/* AND(val, 0x80000000) gives the sign bit.
*
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);
- inst->saturate = true;
- }
} else {
/* For doubles we do the same but we need to consider:
*
set_predicate(BRW_PREDICATE_NORMAL,
bld.OR(r, r, brw_imm_ud(0x3ff00000u)));
-
- if (instr->dest.saturate) {
- inst = bld.MOV(result, result);
- inst->saturate = true;
- }
}
break;
}
- case nir_op_isign: {
- /* ASR(val, 31) -> negative val generates 0xffffffff (signed -1).
- * -> non-negative val generates 0x00000000.
- * Predicated OR sets 1 if val is positive.
- */
- uint32_t bit_size = nir_dest_bit_size(instr->dest.dest);
- assert(bit_size == 32 || bit_size == 16);
-
- fs_reg zero = bit_size == 32 ? brw_imm_d(0) : brw_imm_w(0);
- fs_reg one = bit_size == 32 ? brw_imm_d(1) : brw_imm_w(1);
- fs_reg shift = bit_size == 32 ? brw_imm_d(31) : brw_imm_w(15);
-
- bld.CMP(bld.null_reg_d(), op[0], zero, BRW_CONDITIONAL_G);
- bld.ASR(result, op[0], shift);
- inst = bld.OR(result, result, one);
- inst->predicate = BRW_PREDICATE_NORMAL;
- break;
- }
-
case nir_op_frcp:
inst = bld.emit(SHADER_OPCODE_RCP, result, op[0]);
inst->saturate = instr->dest.saturate;
inst->saturate = instr->dest.saturate;
break;
+ case nir_op_uadd_sat:
+ inst = bld.ADD(result, op[0], op[1]);
+ inst->saturate = true;
+ break;
+
case nir_op_fmul:
inst = bld.MUL(result, op[0], op[1]);
inst->saturate = instr->dest.saturate;
break;
}
- case nir_op_flt:
- case nir_op_fge:
- case nir_op_feq:
- case nir_op_fne: {
+ case nir_op_flt32:
+ case nir_op_fge32:
+ case nir_op_feq32:
+ case nir_op_fne32: {
fs_reg dest = result;
const uint32_t bit_size = nir_src_bit_size(instr->src[0].src);
brw_conditional_mod cond;
switch (instr->op) {
- case nir_op_flt:
+ case nir_op_flt32:
cond = BRW_CONDITIONAL_L;
break;
- case nir_op_fge:
+ case nir_op_fge32:
cond = BRW_CONDITIONAL_GE;
break;
- case nir_op_feq:
+ case nir_op_feq32:
cond = BRW_CONDITIONAL_Z;
break;
- case nir_op_fne:
+ case nir_op_fne32:
cond = BRW_CONDITIONAL_NZ;
break;
default:
break;
}
- case nir_op_ilt:
- case nir_op_ult:
- case nir_op_ige:
- case nir_op_uge:
- case nir_op_ieq:
- case nir_op_ine: {
+ case nir_op_ilt32:
+ case nir_op_ult32:
+ case nir_op_ige32:
+ case nir_op_uge32:
+ case nir_op_ieq32:
+ case nir_op_ine32: {
fs_reg dest = result;
const uint32_t bit_size = nir_src_bit_size(instr->src[0].src);
brw_conditional_mod cond;
switch (instr->op) {
- case nir_op_ilt:
- case nir_op_ult:
+ case nir_op_ilt32:
+ case nir_op_ult32:
cond = BRW_CONDITIONAL_L;
break;
- case nir_op_ige:
- case nir_op_uge:
+ case nir_op_ige32:
+ case nir_op_uge32:
cond = BRW_CONDITIONAL_GE;
break;
- case nir_op_ieq:
+ case nir_op_ieq32:
cond = BRW_CONDITIONAL_Z;
break;
- case nir_op_ine:
+ case nir_op_ine32:
cond = BRW_CONDITIONAL_NZ;
break;
default:
case nir_op_fdot2:
case nir_op_fdot3:
case nir_op_fdot4:
- case nir_op_ball_fequal2:
- case nir_op_ball_iequal2:
- case nir_op_ball_fequal3:
- case nir_op_ball_iequal3:
- case nir_op_ball_fequal4:
- case nir_op_ball_iequal4:
- case nir_op_bany_fnequal2:
- case nir_op_bany_inequal2:
- case nir_op_bany_fnequal3:
- case nir_op_bany_inequal3:
- case nir_op_bany_fnequal4:
- case nir_op_bany_inequal4:
+ case nir_op_b32all_fequal2:
+ case nir_op_b32all_iequal2:
+ case nir_op_b32all_fequal3:
+ case nir_op_b32all_iequal3:
+ case nir_op_b32all_fequal4:
+ case nir_op_b32all_iequal4:
+ case nir_op_b32any_fnequal2:
+ case nir_op_b32any_inequal2:
+ case nir_op_b32any_fnequal3:
+ case nir_op_b32any_inequal3:
+ case nir_op_b32any_fnequal4:
+ case nir_op_b32any_inequal4:
unreachable("Lowered by nir_lower_alu_reductions");
case nir_op_fnoise1_1:
inst->saturate = instr->dest.saturate;
break;
- case nir_op_b2i:
- case nir_op_b2f:
- bld.MOV(result, negate(op[0]));
- break;
-
- case nir_op_i2b:
- case nir_op_f2b: {
+ case nir_op_i2b32:
+ case nir_op_f2b32: {
uint32_t bit_size = nir_src_bit_size(instr->src[0].src);
if (bit_size == 64) {
/* two-argument instructions can't take 64-bit immediates */
fs_reg zero;
fs_reg tmp;
- if (instr->op == nir_op_f2b) {
+ if (instr->op == nir_op_f2b32) {
zero = vgrf(glsl_type::double_type);
tmp = vgrf(glsl_type::double_type);
bld.MOV(zero, setup_imm_df(bld, 0.0));
} else {
fs_reg zero;
if (bit_size == 32) {
- zero = instr->op == nir_op_f2b ? brw_imm_f(0.0f) : brw_imm_d(0);
+ zero = instr->op == nir_op_f2b32 ? brw_imm_f(0.0f) : brw_imm_d(0);
} else {
assert(bit_size == 16);
- zero = instr->op == nir_op_f2b ?
+ zero = instr->op == nir_op_f2b32 ?
retype(brw_imm_w(0), BRW_REGISTER_TYPE_HF) : brw_imm_w(0);
}
bld.CMP(result, op[0], zero, BRW_CONDITIONAL_NZ);
unreachable("not reached: should be handled by lower_packing_builtins");
case nir_op_unpack_half_2x16_split_x:
- inst = bld.emit(FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X, result, op[0]);
+ inst = bld.emit(BRW_OPCODE_F16TO32, result,
+ subscript(op[0], BRW_REGISTER_TYPE_UW, 0));
inst->saturate = instr->dest.saturate;
break;
case nir_op_unpack_half_2x16_split_y:
- inst = bld.emit(FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y, result, op[0]);
+ inst = bld.emit(BRW_OPCODE_F16TO32, result,
+ subscript(op[0], BRW_REGISTER_TYPE_UW, 1));
inst->saturate = instr->dest.saturate;
break;
unreachable("not reached: should have been lowered");
case nir_op_ishl:
+ bld.SHL(result, op[0], op[1]);
+ break;
case nir_op_ishr:
- case nir_op_ushr: {
- fs_reg shift_count = op[1];
-
- if (devinfo->is_cherryview || gen_device_info_is_9lp(devinfo)) {
- if (op[1].file == VGRF &&
- (result.type == BRW_REGISTER_TYPE_Q ||
- result.type == BRW_REGISTER_TYPE_UQ)) {
- shift_count = fs_reg(VGRF, alloc.allocate(dispatch_width / 4),
- BRW_REGISTER_TYPE_UD);
- shift_count.stride = 2;
- bld.MOV(shift_count, op[1]);
- }
- }
-
- switch (instr->op) {
- case nir_op_ishl:
- bld.SHL(result, op[0], shift_count);
- break;
- case nir_op_ishr:
- bld.ASR(result, op[0], shift_count);
- break;
- case nir_op_ushr:
- bld.SHR(result, op[0], shift_count);
- break;
- default:
- unreachable("not reached");
- }
+ bld.ASR(result, op[0], op[1]);
+ break;
+ case nir_op_ushr:
+ bld.SHR(result, op[0], op[1]);
break;
- }
case nir_op_pack_half_2x16_split:
bld.emit(FS_OPCODE_PACK_HALF_2x16_SPLIT, result, op[0], op[1]);
inst->saturate = instr->dest.saturate;
break;
- case nir_op_bcsel:
+ case nir_op_b32csel:
if (optimize_frontfacing_ternary(instr, result))
return;
case nir_op_extract_u8:
case nir_op_extract_i8: {
- nir_const_value *byte = nir_src_as_const_value(instr->src[1].src);
- assert(byte != NULL);
+ unsigned byte = nir_src_as_uint(instr->src[1].src);
/* The PRMs say:
*
if (instr->op == nir_op_extract_i8) {
/* If we need to sign extend, extract to a word first */
fs_reg w_temp = bld.vgrf(BRW_REGISTER_TYPE_W);
- bld.MOV(w_temp, subscript(op[0], type, byte->u32[0]));
+ bld.MOV(w_temp, subscript(op[0], type, byte));
bld.MOV(result, w_temp);
} else {
/* Otherwise use an AND with 0xff and a word type */
- bld.AND(result, subscript(op[0], type, byte->u32[0] / 2), brw_imm_uw(0xff));
+ bld.AND(result, subscript(op[0], type, byte / 2), brw_imm_uw(0xff));
}
} else {
const brw_reg_type type = brw_int_type(1, instr->op == nir_op_extract_i8);
- bld.MOV(result, subscript(op[0], type, byte->u32[0]));
+ bld.MOV(result, subscript(op[0], type, byte));
}
break;
}
case nir_op_extract_u16:
case nir_op_extract_i16: {
const brw_reg_type type = brw_int_type(2, instr->op == nir_op_extract_i16);
- nir_const_value *word = nir_src_as_const_value(instr->src[1].src);
- assert(word != NULL);
- bld.MOV(result, subscript(op[0], type, word->u32[0]));
+ unsigned word = nir_src_as_uint(instr->src[1].src);
+ bld.MOV(result, subscript(op[0], type, word));
break;
}
fs_reg reg = bld.vgrf(reg_type, instr->def.num_components);
switch (instr->def.bit_size) {
+ case 8:
+ for (unsigned i = 0; i < instr->def.num_components; i++)
+ bld.MOV(offset(reg, bld, i), setup_imm_b(bld, instr->value.i8[i]));
+ break;
+
case 16:
for (unsigned i = 0; i < instr->def.num_components; i++)
bld.MOV(offset(reg, bld, i), brw_imm_w(instr->value.i16[i]));
fs_reg
fs_visitor::get_nir_src_imm(const nir_src &src)
{
- nir_const_value *val = nir_src_as_const_value(src);
assert(nir_src_bit_size(src) == 32);
- return val ? fs_reg(brw_imm_d(val->i32[0])) : get_nir_src(src);
+ return nir_src_is_const(src) ?
+ fs_reg(brw_imm_d(nir_src_as_int(src))) : get_nir_src(src);
}
fs_reg
{
if (dest.is_ssa) {
const brw_reg_type reg_type =
- brw_reg_type_from_bit_size(dest.ssa.bit_size, BRW_REGISTER_TYPE_F);
+ brw_reg_type_from_bit_size(dest.ssa.bit_size,
+ dest.ssa.bit_size == 8 ?
+ BRW_REGISTER_TYPE_D :
+ BRW_REGISTER_TYPE_F);
nir_ssa_values[dest.ssa.index] =
bld.vgrf(reg_type, dest.ssa.num_components);
return nir_ssa_values[dest.ssa.index];
}
}
-fs_reg
-fs_visitor::get_nir_image_deref(const nir_deref_var *deref)
-{
- 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) {
- const nir_deref_array *deref_array = nir_deref_as_array(tail->child);
- assert(tail->child->deref_type == nir_deref_type_array);
- const unsigned size = glsl_get_length(tail->type);
- const unsigned element_size = type_size_scalar(deref_array->deref.type);
- const unsigned base = MIN2(deref_array->base_offset, size - 1);
- image = offset(image, bld, base * element_size);
-
- if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
- fs_reg tmp = vgrf(glsl_type::uint_type);
-
- /* Accessing an invalid surface index with the dataport can result
- * in a hang. According to the spec "if the index used to
- * select an individual element is negative or greater than or
- * equal to the size of the array, the results of the operation
- * are undefined but may not lead to termination" -- which is one
- * of the possible outcomes of the hang. Clamp the index to
- * prevent access outside of the array bounds.
- */
- bld.emit_minmax(tmp, retype(get_nir_src(deref_array->indirect),
- BRW_REGISTER_TYPE_UD),
- brw_imm_ud(size - base - 1), BRW_CONDITIONAL_L);
-
- 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);
- }
- }
- }
-
- 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_visitor::emit_percomp(const fs_builder &bld, const fs_inst &inst,
unsigned wr_mask)
}
}
-/**
- * Get the matching channel register datatype for an image intrinsic of the
- * specified GLSL image type.
- */
-static brw_reg_type
-get_image_base_type(const glsl_type *type)
-{
- switch ((glsl_base_type)type->sampled_type) {
- case GLSL_TYPE_UINT:
- return BRW_REGISTER_TYPE_UD;
- case GLSL_TYPE_INT:
- return BRW_REGISTER_TYPE_D;
- case GLSL_TYPE_FLOAT:
- return BRW_REGISTER_TYPE_F;
- default:
- unreachable("Not reached.");
- }
-}
-
-/**
- * Get the appropriate atomic op for an image atomic intrinsic.
- */
-static unsigned
-get_image_atomic_op(nir_intrinsic_op op, const glsl_type *type)
-{
- switch (op) {
- case nir_intrinsic_image_var_atomic_add:
- return BRW_AOP_ADD;
- case nir_intrinsic_image_var_atomic_min:
- return (get_image_base_type(type) == BRW_REGISTER_TYPE_D ?
- BRW_AOP_IMIN : BRW_AOP_UMIN);
- case nir_intrinsic_image_var_atomic_max:
- return (get_image_base_type(type) == BRW_REGISTER_TYPE_D ?
- BRW_AOP_IMAX : BRW_AOP_UMAX);
- case nir_intrinsic_image_var_atomic_and:
- return BRW_AOP_AND;
- case nir_intrinsic_image_var_atomic_or:
- return BRW_AOP_OR;
- case nir_intrinsic_image_var_atomic_xor:
- return BRW_AOP_XOR;
- case nir_intrinsic_image_var_atomic_exchange:
- return BRW_AOP_MOV;
- case nir_intrinsic_image_var_atomic_comp_swap:
- return BRW_AOP_CMPWR;
- default:
- unreachable("Not reachable.");
- }
-}
-
static fs_inst *
emit_pixel_interpolater_send(const fs_builder &bld,
enum opcode opcode,
{
struct brw_wm_prog_data *wm_prog_data =
brw_wm_prog_data(bld.shader->stage_prog_data);
- fs_inst *inst;
- fs_reg payload;
- int mlen;
- if (src.file == BAD_FILE) {
- /* Dummy payload */
- payload = bld.vgrf(BRW_REGISTER_TYPE_F, 1);
- mlen = 1;
- } else {
- payload = src;
- mlen = 2 * bld.dispatch_width() / 8;
- }
-
- inst = bld.emit(opcode, dst, payload, desc);
- inst->mlen = mlen;
+ fs_inst *inst = bld.emit(opcode, dst, src, desc);
/* 2 floats per slot returned */
inst->size_written = 2 * dst.component_size(inst->exec_size);
inst->pi_noperspective = interpolation == INTERP_MODE_NOPERSPECTIVE;
}
/* Store the control data bits in the message payload and send it. */
- int mlen = 2;
+ unsigned mlen = 2;
if (channel_mask.file != BAD_FILE)
mlen += 4; /* channel masks, plus 3 extra copies of the data */
if (per_slot_offset.file != BAD_FILE)
fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, mlen);
fs_reg *sources = ralloc_array(mem_ctx, fs_reg, mlen);
- int i = 0;
+ unsigned i = 0;
sources[i++] = fs_reg(retype(brw_vec8_grf(1, 0), BRW_REGISTER_TYPE_UD));
if (per_slot_offset.file != BAD_FILE)
sources[i++] = per_slot_offset;
unsigned first_component)
{
struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);
-
- 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;
/* TODO: figure out push input layout for invocations == 1 */
/* TODO: make this work with 64-bit inputs */
if (gs_prog_data->invocations == 1 &&
type_sz(dst.type) <= 4 &&
- 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;
+ nir_src_is_const(offset_src) && nir_src_is_const(vertex_src) &&
+ 4 * (base_offset + nir_src_as_uint(offset_src)) < push_reg_count) {
+ int imm_offset = (base_offset + nir_src_as_uint(offset_src)) * 4 +
+ nir_src_as_uint(vertex_src) * push_reg_count;
for (unsigned i = 0; i < num_components; i++) {
bld.MOV(offset(dst, bld, i),
fs_reg(ATTR, imm_offset + i + first_component, dst.type));
fs_reg icp_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
if (gs_prog_data->invocations == 1) {
- if (vertex_const) {
+ if (nir_src_is_const(vertex_src)) {
/* 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),
+ retype(brw_vec8_grf(first_icp_handle + nir_src_as_uint(vertex_src), 0),
BRW_REGISTER_TYPE_UD);
} else {
/* The vertex index is non-constant. We need to use indirect
} else {
assert(gs_prog_data->invocations > 1);
- if (vertex_const) {
- assert(devinfo->gen >= 9 || vertex_const->i32[0] <= 5);
+ if (nir_src_is_const(vertex_src)) {
+ unsigned vertex = nir_src_as_uint(vertex_src);
+ assert(devinfo->gen >= 9 || vertex <= 5);
bld.MOV(icp_handle,
- retype(brw_vec1_grf(first_icp_handle +
- vertex_const->i32[0] / 8,
- vertex_const->i32[0] % 8),
+ retype(brw_vec1_grf(first_icp_handle + vertex / 8, vertex % 8),
BRW_REGISTER_TYPE_UD));
} else {
/* The vertex index is non-constant. We need to use indirect
}
for (unsigned iter = 0; iter < num_iterations; iter++) {
- if (offset_const) {
+ if (nir_src_is_const(offset_src)) {
/* Constant indexing - use global offset. */
if (first_component != 0) {
unsigned read_components = num_components + first_component;
inst->size_written = num_components *
tmp_dst.component_size(inst->exec_size);
}
- inst->offset = base_offset + offset_const->u32[0];
+ inst->offset = base_offset + nir_src_as_uint(offset_src);
inst->mlen = 1;
} else {
/* Indirect indexing - use per-slot offsets as well. */
if (num_iterations > 1) {
num_components = orig_num_components - 2;
- if(offset_const) {
+ if(nir_src_is_const(offset_src)) {
base_offset++;
} else {
fs_reg new_indirect = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
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) {
+ if (nir_src_is_const(*offset_src)) {
/* 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);
+ assert(nir_src_as_uint(*offset_src) == 0);
return fs_reg();
}
return get_nir_src(*offset_src);
}
-static void
-do_untyped_vector_read(const fs_builder &bld,
- const fs_reg dest,
- const fs_reg surf_index,
- const fs_reg offset_reg,
- unsigned num_components)
-{
- if (type_sz(dest.type) <= 2) {
- assert(dest.stride == 1);
- boolean is_const_offset = offset_reg.file == BRW_IMMEDIATE_VALUE;
-
- if (is_const_offset) {
- uint32_t start = offset_reg.ud & ~3;
- uint32_t end = offset_reg.ud + num_components * type_sz(dest.type);
- end = ALIGN(end, 4);
- assert (end - start <= 16);
-
- /* At this point we have 16-bit component/s that have constant
- * offset aligned to 4-bytes that can be read with untyped_reads.
- * untyped_read message requires 32-bit aligned offsets.
- */
- unsigned first_component = (offset_reg.ud & 3) / type_sz(dest.type);
- unsigned num_components_32bit = (end - start) / 4;
-
- fs_reg read_result =
- emit_untyped_read(bld, surf_index, brw_imm_ud(start),
- 1 /* dims */,
- num_components_32bit,
- BRW_PREDICATE_NONE);
- shuffle_from_32bit_read(bld, dest, read_result, first_component,
- num_components);
- } else {
- fs_reg read_offset = bld.vgrf(BRW_REGISTER_TYPE_UD);
- for (unsigned i = 0; i < num_components; i++) {
- if (i == 0) {
- bld.MOV(read_offset, offset_reg);
- } else {
- bld.ADD(read_offset, offset_reg,
- brw_imm_ud(i * type_sz(dest.type)));
- }
- /* Non constant offsets are not guaranteed to be aligned 32-bits
- * so they are read using one byte_scattered_read message
- * for each component.
- */
- fs_reg read_result =
- emit_byte_scattered_read(bld, surf_index, read_offset,
- 1 /* dims */, 1,
- type_sz(dest.type) * 8 /* bit_size */,
- BRW_PREDICATE_NONE);
- bld.MOV(offset(dest, bld, i),
- subscript (read_result, dest.type, 0));
- }
- }
- } else if (type_sz(dest.type) == 4) {
- fs_reg read_result = emit_untyped_read(bld, surf_index, offset_reg,
- 1 /* dims */,
- num_components,
- BRW_PREDICATE_NONE);
- read_result.type = dest.type;
- for (unsigned i = 0; i < num_components; i++)
- bld.MOV(offset(dest, bld, i), offset(read_result, bld, i));
- } else if (type_sz(dest.type) == 8) {
- /* Reading a dvec, so we need to:
- *
- * 1. Multiply num_components by 2, to account for the fact that we
- * need to read 64-bit components.
- * 2. Shuffle the result of the load to form valid 64-bit elements
- * 3. Emit a second load (for components z/w) if needed.
- */
- fs_reg read_offset = bld.vgrf(BRW_REGISTER_TYPE_UD);
- bld.MOV(read_offset, offset_reg);
-
- int iters = num_components <= 2 ? 1 : 2;
-
- /* Load the dvec, the first iteration loads components x/y, the second
- * iteration, if needed, loads components z/w
- */
- for (int it = 0; it < iters; it++) {
- /* Compute number of components to read in this iteration */
- int iter_components = MIN2(2, num_components);
- num_components -= iter_components;
-
- /* Read. Since this message reads 32-bit components, we need to
- * read twice as many components.
- */
- fs_reg read_result = emit_untyped_read(bld, surf_index, read_offset,
- 1 /* dims */,
- iter_components * 2,
- BRW_PREDICATE_NONE);
-
- /* Shuffle the 32-bit load result into valid 64-bit data */
- shuffle_from_32bit_read(bld, offset(dest, bld, it * 2),
- read_result, 0, iter_components);
-
- bld.ADD(read_offset, read_offset, brw_imm_ud(16));
- }
- } else {
- unreachable("Unsupported type");
- }
-}
-
void
fs_visitor::nir_emit_vs_intrinsic(const fs_builder &bld,
nir_intrinsic_instr *instr)
case nir_intrinsic_load_base_vertex:
unreachable("should be lowered by nir_lower_system_values()");
- case nir_intrinsic_load_vertex_id_zero_base:
- 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);
- dest.type = val.type;
- bld.MOV(dest, val);
- break;
- }
-
case nir_intrinsic_load_input: {
fs_reg src = fs_reg(ATTR, nir_intrinsic_base(instr) * 4, dest.type);
unsigned first_component = nir_intrinsic_component(instr);
unsigned num_components = instr->num_components;
- 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]);
+ src = offset(src, bld, nir_src_as_uint(instr->src[0]));
if (type_sz(dest.type) == 8)
first_component /= 2;
- for (unsigned j = 0; j < num_components; j++) {
- bld.MOV(offset(dest, bld, j), offset(src, bld, j + first_component));
- }
-
- if (type_sz(dest.type) == 8) {
- shuffle_32bit_load_result_to_64bit_data(bld,
- dest,
- retype(dest, BRW_REGISTER_TYPE_F),
- instr->num_components);
- }
+ /* For 16-bit support maybe a temporary will be needed to copy from
+ * the ATTR file.
+ */
+ shuffle_from_32bit_read(bld, dest, retype(src, BRW_REGISTER_TYPE_D),
+ first_component, num_components);
break;
}
+ case nir_intrinsic_load_vertex_id_zero_base:
+ case nir_intrinsic_load_instance_id:
+ case nir_intrinsic_load_base_instance:
+ case nir_intrinsic_load_draw_id:
case nir_intrinsic_load_first_vertex:
case nir_intrinsic_load_is_indexed_draw:
unreachable("lowered by brw_nir_lower_vs_inputs");
unsigned imm_offset = instr->const_index[0];
const nir_src &vertex_src = instr->src[0];
- nir_const_value *vertex_const = nir_src_as_const_value(vertex_src);
fs_inst *inst;
fs_reg icp_handle;
- if (vertex_const) {
+ if (nir_src_is_const(vertex_src)) {
/* Emit a MOV to resolve <0,1,0> regioning. */
icp_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
+ unsigned vertex = nir_src_as_uint(vertex_src);
bld.MOV(icp_handle,
- retype(brw_vec1_grf(1 + (vertex_const->i32[0] >> 3),
- vertex_const->i32[0] & 7),
+ retype(brw_vec1_grf(1 + (vertex >> 3), vertex & 7),
BRW_REGISTER_TYPE_UD));
} else if (tcs_prog_data->instances == 1 &&
vertex_src.is_ssa &&
* or SSBOs.
*/
if (type_sz(dst.type) == 8) {
- shuffle_32bit_load_result_to_64bit_data(
- bld, dst, retype(dst, BRW_REGISTER_TYPE_F), num_components);
-
- for (unsigned c = 0; c < num_components; c++) {
- bld.MOV(offset(orig_dst, bld, iter * 2 + c),
- offset(dst, bld, c));
- }
+ shuffle_from_32bit_read(bld,
+ offset(orig_dst, bld, iter * 2),
+ retype(dst, BRW_REGISTER_TYPE_D),
+ 0, num_components);
}
/* Copy the temporary to the destination to deal with writemasking.
* for that.
*/
unsigned channel = iter * 2 + i;
- fs_reg dest = shuffle_64bit_data_for_32bit_write(bld,
- offset(value, bld, channel), 1);
+ fs_reg dest = shuffle_for_32bit_write(bld, value, channel, 1);
srcs[header_regs + (i + first_component) * 2] = dest;
srcs[header_regs + (i + first_component) * 2 + 1] =
* or SSBOs.
*/
if (type_sz(dest.type) == 8) {
- shuffle_32bit_load_result_to_64bit_data(
- bld, dest, retype(dest, BRW_REGISTER_TYPE_F), num_components);
-
- for (unsigned c = 0; c < num_components; c++) {
- bld.MOV(offset(orig_dest, bld, iter * 2 + c),
- offset(dest, bld, c));
- }
+ shuffle_from_32bit_read(bld,
+ offset(orig_dest, bld, iter * 2),
+ retype(dest, BRW_REGISTER_TYPE_D),
+ 0, num_components);
}
/* If we are loading double data and we need a second read message
wm_prog_data->binding_table.render_target_read_start -
wm_prog_data->base.binding_table.texture_start;
- brw_mark_surface_used(
- bld.shader->stage_prog_data,
- wm_prog_data->binding_table.render_target_read_start + target);
-
/* Calculate the fragment coordinates. */
const fs_reg coords = bld.vgrf(BRW_REGISTER_TYPE_UD, 3);
bld.MOV(offset(coords, bld, 0), pixel_x);
SHADER_OPCODE_TXF_CMS_LOGICAL;
/* Emit the instruction. */
- const fs_reg srcs[] = { coords, fs_reg(), brw_imm_ud(0), fs_reg(),
- sample, mcs,
- brw_imm_ud(surface), brw_imm_ud(0),
- fs_reg(), brw_imm_ud(3), brw_imm_ud(0) };
- STATIC_ASSERT(ARRAY_SIZE(srcs) == TEX_LOGICAL_NUM_SRCS);
+ fs_reg srcs[TEX_LOGICAL_NUM_SRCS];
+ srcs[TEX_LOGICAL_SRC_COORDINATE] = coords;
+ srcs[TEX_LOGICAL_SRC_LOD] = brw_imm_ud(0);
+ srcs[TEX_LOGICAL_SRC_SAMPLE_INDEX] = sample;
+ srcs[TEX_LOGICAL_SRC_MCS] = mcs;
+ srcs[TEX_LOGICAL_SRC_SURFACE] = brw_imm_ud(surface);
+ srcs[TEX_LOGICAL_SRC_SAMPLER] = brw_imm_ud(0);
+ srcs[TEX_LOGICAL_SRC_COORD_COMPONENTS] = brw_imm_ud(3);
+ srcs[TEX_LOGICAL_SRC_GRAD_COMPONENTS] = brw_imm_ud(0);
fs_inst *inst = bld.emit(op, dst, srcs, ARRAY_SIZE(srcs));
inst->size_written = 4 * inst->dst.component_size(inst->exec_size);
case nir_intrinsic_store_output: {
const fs_reg src = get_nir_src(instr->src[0]);
- const nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
- assert(const_offset && "Indirect output stores not allowed");
+ const unsigned store_offset = nir_src_as_uint(instr->src[1]);
const unsigned location = nir_intrinsic_base(instr) +
- SET_FIELD(const_offset->u32[0], BRW_NIR_FRAG_OUTPUT_LOCATION);
+ SET_FIELD(store_offset, BRW_NIR_FRAG_OUTPUT_LOCATION);
const fs_reg new_dest = retype(alloc_frag_output(this, location),
src.type);
const unsigned l = GET_FIELD(nir_intrinsic_base(instr),
BRW_NIR_FRAG_OUTPUT_LOCATION);
assert(l >= FRAG_RESULT_DATA0);
- nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
- assert(const_offset && "Indirect output loads not allowed");
- const unsigned target = l - FRAG_RESULT_DATA0 + const_offset->u32[0];
+ const unsigned load_offset = nir_src_as_uint(instr->src[0]);
+ const unsigned target = l - FRAG_RESULT_DATA0 + load_offset;
const fs_reg tmp = bld.vgrf(dest.type, 4);
if (reinterpret_cast<const brw_wm_prog_key *>(key)->coherent_fb_fetch)
if (devinfo->gen >= 6) {
emit_discard_jump();
}
+
+ limit_dispatch_width(16, "Fragment discard not implemented in SIMD32 mode.");
break;
}
unsigned base = nir_intrinsic_base(instr);
unsigned comp = nir_intrinsic_component(instr);
unsigned num_components = instr->num_components;
+ fs_reg orig_dest = dest;
enum brw_reg_type type = dest.type;
/* Special case fields in the VUE header */
*/
type = BRW_REGISTER_TYPE_F;
num_components *= 2;
+ dest = bld.vgrf(type, num_components);
}
for (unsigned int i = 0; i < num_components; i++) {
}
if (nir_dest_bit_size(instr->dest) == 64) {
- shuffle_32bit_load_result_to_64bit_data(bld,
- dest,
- retype(dest, type),
- instr->num_components);
+ shuffle_from_32bit_read(bld, orig_dest, dest, 0,
+ instr->num_components);
}
break;
}
const glsl_interp_mode interpolation =
(enum glsl_interp_mode) nir_intrinsic_interp_mode(instr);
- nir_const_value *const_sample = nir_src_as_const_value(instr->src[0]);
-
- if (const_sample) {
- unsigned msg_data = const_sample->i32[0] << 4;
+ if (nir_src_is_const(instr->src[0])) {
+ unsigned msg_data = nir_src_as_uint(instr->src[0]) << 4;
emit_pixel_interpolater_send(bld,
FS_OPCODE_INTERPOLATE_AT_SAMPLE,
FS_OPCODE_INTERPOLATE_AT_SAMPLE,
dest,
fs_reg(), /* src */
- msg_data,
+ component(msg_data, 0),
interpolation);
set_predicate(BRW_PREDICATE_NORMAL, inst);
nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
if (const_offset) {
+ assert(nir_src_bit_size(instr->src[0]) == 32);
unsigned off_x = MIN2((int)(const_offset->f32[0] * 16), 7) & 0xf;
unsigned off_y = MIN2((int)(const_offset->f32[1] * 16), 7) & 0xf;
}
}
+static int
+get_op_for_atomic_add(nir_intrinsic_instr *instr, unsigned src)
+{
+ if (nir_src_is_const(instr->src[src])) {
+ int64_t add_val = nir_src_as_int(instr->src[src]);
+ if (add_val == 1)
+ return BRW_AOP_INC;
+ else if (add_val == -1)
+ return BRW_AOP_DEC;
+ }
+
+ return BRW_AOP_ADD;
+}
+
void
fs_visitor::nir_emit_cs_intrinsic(const fs_builder &bld,
nir_intrinsic_instr *instr)
cs_prog_data->uses_num_work_groups = true;
- fs_reg surf_index = brw_imm_ud(surface);
- brw_mark_surface_used(prog_data, surface);
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] = brw_imm_ud(surface);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(1); /* num components */
/* 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,
- brw_imm_ud(i << 2),
- 1 /* dims */, 1 /* size */,
- BRW_PREDICATE_NONE);
- read_result.type = dest.type;
- bld.MOV(dest, read_result);
- dest = offset(dest, bld, 1);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = brw_imm_ud(i << 2);
+ bld.emit(SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL,
+ offset(dest, bld, i), srcs, SURFACE_LOGICAL_NUM_SRCS);
}
break;
}
case nir_intrinsic_shared_atomic_add:
- nir_emit_shared_atomic(bld, BRW_AOP_ADD, instr);
+ nir_emit_shared_atomic(bld, get_op_for_atomic_add(instr, 1), instr);
break;
case nir_intrinsic_shared_atomic_imin:
nir_emit_shared_atomic(bld, BRW_AOP_IMIN, instr);
case nir_intrinsic_shared_atomic_comp_swap:
nir_emit_shared_atomic(bld, BRW_AOP_CMPWR, instr);
break;
+ case nir_intrinsic_shared_atomic_fmin:
+ nir_emit_shared_atomic_float(bld, BRW_AOP_FMIN, instr);
+ break;
+ case nir_intrinsic_shared_atomic_fmax:
+ nir_emit_shared_atomic_float(bld, BRW_AOP_FMAX, instr);
+ break;
+ case nir_intrinsic_shared_atomic_fcomp_swap:
+ nir_emit_shared_atomic_float(bld, BRW_AOP_FCMPWR, instr);
+ break;
case nir_intrinsic_load_shared: {
assert(devinfo->gen >= 7);
+ assert(stage == MESA_SHADER_COMPUTE);
- fs_reg surf_index = brw_imm_ud(GEN7_BTI_SLM);
+ const unsigned bit_size = nir_dest_bit_size(instr->dest);
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] = brw_imm_ud(GEN7_BTI_SLM);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[0]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
- /* 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]));
- }
+ /* Make dest unsigned because that's what the temporary will be */
+ dest.type = brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_UD);
/* Read the vector */
- do_untyped_vector_read(bld, dest, surf_index, offset_reg,
- instr->num_components);
+ if (nir_intrinsic_align(instr) >= 4) {
+ assert(nir_dest_bit_size(instr->dest) == 32);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(instr->num_components);
+ fs_inst *inst =
+ bld.emit(SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
+ inst->size_written = instr->num_components * dispatch_width * 4;
+ } else {
+ assert(nir_dest_bit_size(instr->dest) <= 32);
+ assert(nir_dest_num_components(instr->dest) == 1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(bit_size);
+
+ fs_reg read_result = bld.vgrf(BRW_REGISTER_TYPE_UD);
+ bld.emit(SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL,
+ read_result, srcs, SURFACE_LOGICAL_NUM_SRCS);
+ bld.MOV(dest, read_result);
+ }
break;
}
case nir_intrinsic_store_shared: {
assert(devinfo->gen >= 7);
+ assert(stage == MESA_SHADER_COMPUTE);
+
+ const unsigned bit_size = nir_src_bit_size(instr->src[0]);
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] = brw_imm_ud(GEN7_BTI_SLM);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[1]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+
+ fs_reg data = get_nir_src(instr->src[0]);
+ data.type = brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_UD);
+
+ assert(nir_intrinsic_write_mask(instr) ==
+ (1u << instr->num_components) - 1);
+ if (nir_intrinsic_align(instr) >= 4) {
+ assert(nir_src_bit_size(instr->src[0]) == 32);
+ assert(nir_src_num_components(instr->src[0]) <= 4);
+ srcs[SURFACE_LOGICAL_SRC_DATA] = data;
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(instr->num_components);
+ bld.emit(SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL,
+ fs_reg(), srcs, SURFACE_LOGICAL_NUM_SRCS);
+ } else {
+ assert(nir_src_bit_size(instr->src[0]) <= 32);
+ assert(nir_src_num_components(instr->src[0]) == 1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(bit_size);
- /* 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];
-
- /* get_nir_src() retypes to integer. Be wary of 64-bit types though
- * since the untyped writes below operate in units of 32-bits, which
- * means that we need to write twice as many components each time.
- * Also, we have to suffle 64-bit data to be in the appropriate layout
- * expected by our 32-bit write messages.
- */
- unsigned type_size = 4;
- if (nir_src_bit_size(instr->src[0]) == 64) {
- type_size = 8;
- val_reg = shuffle_64bit_data_for_32bit_write(bld,
- val_reg, instr->num_components);
- }
-
- unsigned type_slots = type_size / 4;
-
- /* 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;
-
- /* We can't write more than 2 64-bit components at once. Limit the
- * length of the write to what we can do and let the next iteration
- * handle the rest
- */
- if (type_size > 4)
- length = MIN2(2, length);
-
- 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] +
- type_size * 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] + type_size * first_component));
- }
-
- emit_untyped_write(bld, surf_index, offset_reg,
- offset(val_reg, bld, first_component * type_slots),
- 1 /* dims */, length * type_slots,
- BRW_PREDICATE_NONE);
+ srcs[SURFACE_LOGICAL_SRC_DATA] = bld.vgrf(BRW_REGISTER_TYPE_UD);
+ bld.MOV(srcs[SURFACE_LOGICAL_SRC_DATA], data);
- /* 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));
+ bld.emit(SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL,
+ fs_reg(), srcs, SURFACE_LOGICAL_NUM_SRCS);
}
-
break;
}
}
}
+fs_reg
+fs_visitor::get_nir_image_intrinsic_image(const brw::fs_builder &bld,
+ nir_intrinsic_instr *instr)
+{
+ fs_reg image = retype(get_nir_src_imm(instr->src[0]), BRW_REGISTER_TYPE_UD);
+
+ if (stage_prog_data->binding_table.image_start > 0) {
+ if (image.file == BRW_IMMEDIATE_VALUE) {
+ image.d += stage_prog_data->binding_table.image_start;
+ } else {
+ bld.ADD(image, image,
+ brw_imm_d(stage_prog_data->binding_table.image_start));
+ }
+ }
+
+ return bld.emit_uniformize(image);
+}
+
+fs_reg
+fs_visitor::get_nir_ssbo_intrinsic_index(const brw::fs_builder &bld,
+ nir_intrinsic_instr *instr)
+{
+ /* SSBO stores are weird in that their index is in src[1] */
+ const unsigned src = instr->intrinsic == nir_intrinsic_store_ssbo ? 1 : 0;
+
+ fs_reg surf_index;
+ if (nir_src_is_const(instr->src[src])) {
+ unsigned index = stage_prog_data->binding_table.ssbo_start +
+ nir_src_as_uint(instr->src[src]);
+ surf_index = brw_imm_ud(index);
+ } else {
+ surf_index = vgrf(glsl_type::uint_type);
+ bld.ADD(surf_index, get_nir_src(instr->src[src]),
+ brw_imm_ud(stage_prog_data->binding_table.ssbo_start));
+ }
+
+ return bld.emit_uniformize(surf_index);
+}
+
+static unsigned
+image_intrinsic_coord_components(nir_intrinsic_instr *instr)
+{
+ switch (nir_intrinsic_image_dim(instr)) {
+ case GLSL_SAMPLER_DIM_1D:
+ return 1 + nir_intrinsic_image_array(instr);
+ case GLSL_SAMPLER_DIM_2D:
+ case GLSL_SAMPLER_DIM_RECT:
+ return 2 + nir_intrinsic_image_array(instr);
+ case GLSL_SAMPLER_DIM_3D:
+ case GLSL_SAMPLER_DIM_CUBE:
+ return 3;
+ case GLSL_SAMPLER_DIM_BUF:
+ return 1;
+ case GLSL_SAMPLER_DIM_MS:
+ return 2 + nir_intrinsic_image_array(instr);
+ default:
+ unreachable("Invalid image dimension");
+ }
+}
+
void
fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr)
{
dest = get_nir_dest(instr->dest);
switch (instr->intrinsic) {
- case nir_intrinsic_image_var_load:
- case nir_intrinsic_image_var_store:
- case nir_intrinsic_image_var_atomic_add:
- case nir_intrinsic_image_var_atomic_min:
- case nir_intrinsic_image_var_atomic_max:
- case nir_intrinsic_image_var_atomic_and:
- case nir_intrinsic_image_var_atomic_or:
- case nir_intrinsic_image_var_atomic_xor:
- case nir_intrinsic_image_var_atomic_exchange:
- case nir_intrinsic_image_var_atomic_comp_swap: {
- using namespace image_access;
-
+ case nir_intrinsic_image_load:
+ case nir_intrinsic_image_store:
+ case nir_intrinsic_image_atomic_add:
+ case nir_intrinsic_image_atomic_min:
+ case nir_intrinsic_image_atomic_max:
+ case nir_intrinsic_image_atomic_and:
+ case nir_intrinsic_image_atomic_or:
+ case nir_intrinsic_image_atomic_xor:
+ case nir_intrinsic_image_atomic_exchange:
+ case nir_intrinsic_image_atomic_comp_swap: {
if (stage == MESA_SHADER_FRAGMENT &&
- instr->intrinsic != nir_intrinsic_image_var_load)
+ instr->intrinsic != nir_intrinsic_image_load)
brw_wm_prog_data(prog_data)->has_side_effects = true;
- /* Get the referenced image variable and type. */
- const nir_variable *var = instr->variables[0]->var;
- const glsl_type *type = var->type->without_array();
- const brw_reg_type base_type = get_image_base_type(type);
-
/* Get some metadata from the image intrinsic. */
const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic];
- const unsigned arr_dims = type->sampler_array ? 1 : 0;
- const unsigned surf_dims = type->coordinate_components() - arr_dims;
- const unsigned format = var->data.image.format;
- const unsigned dest_components = nir_intrinsic_dest_components(instr);
-
- /* Get the arguments of the image intrinsic. */
- const fs_reg image = get_nir_image_deref(instr->variables[0]);
- const fs_reg addr = retype(get_nir_src(instr->src[0]),
- BRW_REGISTER_TYPE_UD);
- const fs_reg src0 = (info->num_srcs >= 3 ?
- retype(get_nir_src(instr->src[2]), base_type) :
- fs_reg());
- const fs_reg src1 = (info->num_srcs >= 4 ?
- retype(get_nir_src(instr->src[3]), base_type) :
- fs_reg());
- fs_reg tmp;
+ const GLenum format = nir_intrinsic_format(instr);
+
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] =
+ get_nir_image_intrinsic_image(bld, instr);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[1]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] =
+ brw_imm_ud(image_intrinsic_coord_components(instr));
/* Emit an image load, store or atomic op. */
- if (instr->intrinsic == nir_intrinsic_image_var_load)
- tmp = emit_image_load(bld, image, addr, surf_dims, arr_dims, format);
+ if (instr->intrinsic == nir_intrinsic_image_load) {
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(instr->num_components);
+ fs_inst *inst =
+ bld.emit(SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
+ inst->size_written = instr->num_components * dispatch_width * 4;
+ } else if (instr->intrinsic == nir_intrinsic_image_store) {
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(instr->num_components);
+ srcs[SURFACE_LOGICAL_SRC_DATA] = get_nir_src(instr->src[3]);
+ bld.emit(SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL,
+ fs_reg(), srcs, SURFACE_LOGICAL_NUM_SRCS);
+ } else {
+ int op;
+ unsigned num_srcs = info->num_srcs;
- else if (instr->intrinsic == nir_intrinsic_image_var_store)
- emit_image_store(bld, image, addr, src0, surf_dims, arr_dims,
- var->data.image.write_only ? GL_NONE : format);
+ switch (instr->intrinsic) {
+ case nir_intrinsic_image_atomic_add:
+ assert(num_srcs == 4);
- else
- tmp = emit_image_atomic(bld, image, addr, src0, src1,
- surf_dims, arr_dims, dest_components,
- get_image_atomic_op(instr->intrinsic, type));
-
- /* Assign the result. */
- for (unsigned c = 0; c < dest_components; ++c) {
- bld.MOV(offset(retype(dest, base_type), bld, c),
- offset(tmp, bld, c));
+ op = get_op_for_atomic_add(instr, 3);
+
+ if (op != BRW_AOP_ADD)
+ num_srcs = 3;
+ break;
+ case nir_intrinsic_image_atomic_min:
+ assert(format == GL_R32UI || format == GL_R32I);
+ op = (format == GL_R32I) ? BRW_AOP_IMIN : BRW_AOP_UMIN;
+ break;
+ case nir_intrinsic_image_atomic_max:
+ assert(format == GL_R32UI || format == GL_R32I);
+ op = (format == GL_R32I) ? BRW_AOP_IMAX : BRW_AOP_UMAX;
+ break;
+ case nir_intrinsic_image_atomic_and:
+ op = BRW_AOP_AND;
+ break;
+ case nir_intrinsic_image_atomic_or:
+ op = BRW_AOP_OR;
+ break;
+ case nir_intrinsic_image_atomic_xor:
+ op = BRW_AOP_XOR;
+ break;
+ case nir_intrinsic_image_atomic_exchange:
+ op = BRW_AOP_MOV;
+ break;
+ case nir_intrinsic_image_atomic_comp_swap:
+ op = BRW_AOP_CMPWR;
+ break;
+ default:
+ unreachable("Not reachable.");
+ }
+
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(op);
+
+ fs_reg data;
+ if (num_srcs >= 4)
+ data = get_nir_src(instr->src[3]);
+ if (num_srcs >= 5) {
+ fs_reg tmp = bld.vgrf(data.type, 2);
+ fs_reg sources[2] = { data, get_nir_src(instr->src[4]) };
+ bld.LOAD_PAYLOAD(tmp, sources, 2, 0);
+ data = tmp;
+ }
+ srcs[SURFACE_LOGICAL_SRC_DATA] = data;
+
+ bld.emit(SHADER_OPCODE_TYPED_ATOMIC_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
}
break;
}
+ case nir_intrinsic_image_size: {
+ /* Unlike the [un]typed load and store opcodes, the TXS that this turns
+ * into will handle the binding table index for us in the geneerator.
+ */
+ fs_reg image = retype(get_nir_src_imm(instr->src[0]),
+ BRW_REGISTER_TYPE_UD);
+ image = bld.emit_uniformize(image);
+
+ fs_reg srcs[TEX_LOGICAL_NUM_SRCS];
+ srcs[TEX_LOGICAL_SRC_SURFACE] = image;
+ srcs[TEX_LOGICAL_SRC_SAMPLER] = brw_imm_d(0);
+ srcs[TEX_LOGICAL_SRC_COORD_COMPONENTS] = brw_imm_d(0);
+ srcs[TEX_LOGICAL_SRC_GRAD_COMPONENTS] = brw_imm_d(0);
+
+ /* Since the image size is always uniform, we can just emit a SIMD8
+ * query instruction and splat the result out.
+ */
+ const fs_builder ubld = bld.exec_all().group(8, 0);
+
+ fs_reg tmp = ubld.vgrf(BRW_REGISTER_TYPE_UD, 4);
+ fs_inst *inst = ubld.emit(SHADER_OPCODE_IMAGE_SIZE_LOGICAL,
+ tmp, srcs, ARRAY_SIZE(srcs));
+ inst->size_written = 4 * REG_SIZE;
+
+ for (unsigned c = 0; c < instr->dest.ssa.num_components; ++c) {
+ if (c == 2 && nir_intrinsic_image_dim(instr) == GLSL_SAMPLER_DIM_CUBE) {
+ bld.emit(SHADER_OPCODE_INT_QUOTIENT,
+ offset(retype(dest, tmp.type), bld, c),
+ component(offset(tmp, ubld, c), 0), brw_imm_ud(6));
+ } else {
+ bld.MOV(offset(retype(dest, tmp.type), bld, c),
+ component(offset(tmp, ubld, c), 0));
+ }
+ }
+ break;
+ }
+
+ case nir_intrinsic_image_load_raw_intel: {
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] =
+ get_nir_image_intrinsic_image(bld, instr);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[1]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(instr->num_components);
+
+ fs_inst *inst =
+ bld.emit(SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
+ inst->size_written = instr->num_components * dispatch_width * 4;
+ break;
+ }
+
+ case nir_intrinsic_image_store_raw_intel: {
+ if (stage == MESA_SHADER_FRAGMENT)
+ brw_wm_prog_data(prog_data)->has_side_effects = true;
+
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] =
+ get_nir_image_intrinsic_image(bld, instr);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[1]);
+ srcs[SURFACE_LOGICAL_SRC_DATA] = get_nir_src(instr->src[2]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(instr->num_components);
+
+ bld.emit(SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL,
+ fs_reg(), srcs, SURFACE_LOGICAL_NUM_SRCS);
+ break;
+ }
+
+ case nir_intrinsic_group_memory_barrier:
+ case nir_intrinsic_memory_barrier_shared:
case nir_intrinsic_memory_barrier_atomic_counter:
case nir_intrinsic_memory_barrier_buffer:
case nir_intrinsic_memory_barrier_image:
break;
}
- case nir_intrinsic_group_memory_barrier:
- case nir_intrinsic_memory_barrier_shared:
- /* We treat these workgroup-level barriers as no-ops. This should be
- * safe at present and as long as:
- *
- * - Memory access instructions are not subsequently reordered by the
- * compiler back-end.
- *
- * - All threads from a given compute shader workgroup fit within a
- * single subslice and therefore talk to the same HDC shared unit
- * what supposedly guarantees ordering and coherency between threads
- * from the same workgroup. This may change in the future when we
- * start splitting workgroups across multiple subslices.
- *
- * - The context is not in fault-and-stream mode, which could cause
- * memory transactions (including to SLM) prior to the barrier to be
- * replayed after the barrier if a pagefault occurs. This shouldn't
- * be a problem up to and including SKL because fault-and-stream is
- * not usable due to hardware issues, but that's likely to change in
- * the future.
- */
- break;
-
case nir_intrinsic_shader_clock: {
/* We cannot do anything if there is an event, so ignore it for now */
const fs_reg shader_clock = get_timestamp(bld);
break;
}
- case nir_intrinsic_image_var_size: {
- /* Get the referenced image variable and type. */
- const nir_variable *var = instr->variables[0]->var;
- const glsl_type *type = var->type->without_array();
-
- /* Get the size of the image. */
- const fs_reg image = get_nir_image_deref(instr->variables[0]);
- const fs_reg size = offset(image, bld, BRW_IMAGE_PARAM_SIZE_OFFSET);
-
- /* For 1DArray image types, the array index is stored in the Z component.
- * Fix this by swizzling the Z component to the Y component.
- */
- const bool is_1d_array_image =
- type->sampler_dimensionality == GLSL_SAMPLER_DIM_1D &&
- type->sampler_array;
-
- /* For CubeArray images, we should count the number of cubes instead
- * of the number of faces. Fix it by dividing the (Z component) by 6.
- */
- const bool is_cube_array_image =
- type->sampler_dimensionality == GLSL_SAMPLER_DIM_CUBE &&
- type->sampler_array;
-
- /* Copy all the components. */
- for (unsigned c = 0; c < instr->dest.ssa.num_components; ++c) {
- if ((int)c >= type->coordinate_components()) {
- bld.MOV(offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
- 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), brw_imm_d(6));
- } else {
- bld.MOV(offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
- offset(size, bld, c));
- }
- }
-
- break;
- }
-
- case nir_intrinsic_image_var_samples:
+ case nir_intrinsic_image_samples:
/* The driver does not support multi-sampled images. */
bld.MOV(retype(dest, BRW_REGISTER_TYPE_D), brw_imm_d(1));
break;
fs_reg src(UNIFORM, instr->const_index[0] / 4, dest.type);
- nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
- if (const_offset) {
- assert(const_offset->u32[0] % type_sz(dest.type) == 0);
+ if (nir_src_is_const(instr->src[0])) {
+ unsigned load_offset = nir_src_as_uint(instr->src[0]);
+ assert(load_offset % type_sz(dest.type) == 0);
/* For 16-bit types we add the module of the const_index[0]
* offset to access to not 32-bit aligned element
*/
- src.offset = const_offset->u32[0] + instr->const_index[0] % 4;
+ src.offset = load_offset + instr->const_index[0] % 4;
for (unsigned j = 0; j < instr->num_components; j++) {
bld.MOV(offset(dest, bld, j), offset(src, bld, j));
}
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) {
+ if (nir_src_is_const(instr->src[0])) {
const unsigned index = stage_prog_data->binding_table.ubo_start +
- const_index->u32[0];
+ nir_src_as_uint(instr->src[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
* per-channel and add the base UBO index; we have to select a value
bld.ADD(surf_index, get_nir_src(instr->src[0]),
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
- * a tighter bound, but the array information is already lowered away.
- */
- brw_mark_surface_used(prog_data,
- stage_prog_data->binding_table.ubo_start +
- nir->info.num_ubos - 1);
}
- nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
- if (const_offset == NULL) {
+ if (!nir_src_is_const(instr->src[1])) {
fs_reg base_offset = retype(get_nir_src(instr->src[1]),
BRW_REGISTER_TYPE_UD);
* and we have to split it if necessary.
*/
const unsigned type_size = type_sz(dest.type);
+ const unsigned load_offset = nir_src_as_uint(instr->src[1]);
/* See if we've selected this as a push constant candidate */
- if (const_index) {
- const unsigned ubo_block = const_index->u32[0];
- const unsigned offset_256b = const_offset->u32[0] / 32;
+ if (nir_src_is_const(instr->src[0])) {
+ const unsigned ubo_block = nir_src_as_uint(instr->src[0]);
+ const unsigned offset_256b = load_offset / 32;
fs_reg push_reg;
for (int i = 0; i < 4; i++) {
offset_256b < range->start + range->length) {
push_reg = fs_reg(UNIFORM, UBO_START + i, dest.type);
- push_reg.offset = const_offset->u32[0] - 32 * range->start;
+ push_reg.offset = load_offset - 32 * range->start;
break;
}
}
const fs_reg packed_consts = ubld.vgrf(BRW_REGISTER_TYPE_UD);
for (unsigned c = 0; c < instr->num_components;) {
- const unsigned base = const_offset->u32[0] + c * type_size;
+ const unsigned base = load_offset + c * type_size;
/* Number of usable components in the next block-aligned load. */
const unsigned count = MIN2(instr->num_components - c,
(block_sz - base % block_sz) / type_size);
break;
}
- case nir_intrinsic_load_ssbo: {
- assert(devinfo->gen >= 7);
-
- nir_const_value *const_uniform_block =
- nir_src_as_const_value(instr->src[0]);
-
- fs_reg surf_index;
- if (const_uniform_block) {
- unsigned index = stage_prog_data->binding_table.ssbo_start +
- 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]),
- 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.
- */
- brw_mark_surface_used(prog_data,
- stage_prog_data->binding_table.ssbo_start +
- nir->info.num_ssbos - 1);
- }
+ case nir_intrinsic_load_global: {
+ assert(devinfo->gen >= 8);
- 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(const_offset->u32[0]);
+ if (nir_intrinsic_align(instr) >= 4) {
+ assert(nir_dest_bit_size(instr->dest) == 32);
+ fs_inst *inst = bld.emit(SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL,
+ dest,
+ get_nir_src(instr->src[0]), /* Address */
+ fs_reg(), /* No source data */
+ brw_imm_ud(instr->num_components));
+ inst->size_written = instr->num_components *
+ inst->dst.component_size(inst->exec_size);
} else {
- offset_reg = retype(get_nir_src(instr->src[1]), BRW_REGISTER_TYPE_UD);
+ const unsigned bit_size = nir_dest_bit_size(instr->dest);
+ assert(bit_size <= 32);
+ assert(nir_dest_num_components(instr->dest) == 1);
+ brw_reg_type data_type =
+ brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_UD);
+ fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD);
+ bld.emit(SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL,
+ tmp,
+ get_nir_src(instr->src[0]), /* Address */
+ fs_reg(), /* No source data */
+ brw_imm_ud(bit_size));
+ bld.MOV(retype(dest, data_type), tmp);
}
-
- /* Read the vector */
- do_untyped_vector_read(bld, dest, surf_index, offset_reg,
- instr->num_components);
-
break;
}
- case nir_intrinsic_store_ssbo: {
- assert(devinfo->gen >= 7);
+ case nir_intrinsic_store_global:
+ assert(devinfo->gen >= 8);
if (stage == MESA_SHADER_FRAGMENT)
brw_wm_prog_data(prog_data)->has_side_effects = true;
- /* Block index */
- fs_reg surf_index;
- nir_const_value *const_uniform_block =
- nir_src_as_const_value(instr->src[1]);
- if (const_uniform_block) {
- unsigned index = stage_prog_data->binding_table.ssbo_start +
- const_uniform_block->u32[0];
- surf_index = brw_imm_ud(index);
- brw_mark_surface_used(prog_data, index);
+ if (nir_intrinsic_align(instr) >= 4) {
+ assert(nir_src_bit_size(instr->src[0]) == 32);
+ bld.emit(SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL,
+ fs_reg(),
+ get_nir_src(instr->src[1]), /* Address */
+ get_nir_src(instr->src[0]), /* Data */
+ brw_imm_ud(instr->num_components));
} else {
- surf_index = vgrf(glsl_type::uint_type);
- bld.ADD(surf_index, get_nir_src(instr->src[1]),
- brw_imm_ud(stage_prog_data->binding_table.ssbo_start));
-
- brw_mark_surface_used(prog_data,
- stage_prog_data->binding_table.ssbo_start +
- nir->info.num_ssbos - 1);
+ const unsigned bit_size = nir_src_bit_size(instr->src[0]);
+ assert(bit_size <= 32);
+ assert(nir_src_num_components(instr->src[0]) == 1);
+ brw_reg_type data_type =
+ brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_UD);
+ fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD);
+ bld.MOV(tmp, retype(get_nir_src(instr->src[0]), data_type));
+ bld.emit(SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL,
+ fs_reg(),
+ get_nir_src(instr->src[1]), /* Address */
+ tmp, /* Data */
+ brw_imm_ud(nir_src_bit_size(instr->src[0])));
}
+ break;
- /* Value */
- fs_reg val_reg = get_nir_src(instr->src[0]);
+ case nir_intrinsic_global_atomic_add:
+ nir_emit_global_atomic(bld, get_op_for_atomic_add(instr, 1), instr);
+ break;
+ case nir_intrinsic_global_atomic_imin:
+ nir_emit_global_atomic(bld, BRW_AOP_IMIN, instr);
+ break;
+ case nir_intrinsic_global_atomic_umin:
+ nir_emit_global_atomic(bld, BRW_AOP_UMIN, instr);
+ break;
+ case nir_intrinsic_global_atomic_imax:
+ nir_emit_global_atomic(bld, BRW_AOP_IMAX, instr);
+ break;
+ case nir_intrinsic_global_atomic_umax:
+ nir_emit_global_atomic(bld, BRW_AOP_UMAX, instr);
+ break;
+ case nir_intrinsic_global_atomic_and:
+ nir_emit_global_atomic(bld, BRW_AOP_AND, instr);
+ break;
+ case nir_intrinsic_global_atomic_or:
+ nir_emit_global_atomic(bld, BRW_AOP_OR, instr);
+ break;
+ case nir_intrinsic_global_atomic_xor:
+ nir_emit_global_atomic(bld, BRW_AOP_XOR, instr);
+ break;
+ case nir_intrinsic_global_atomic_exchange:
+ nir_emit_global_atomic(bld, BRW_AOP_MOV, instr);
+ break;
+ case nir_intrinsic_global_atomic_comp_swap:
+ nir_emit_global_atomic(bld, BRW_AOP_CMPWR, instr);
+ break;
+ case nir_intrinsic_global_atomic_fmin:
+ nir_emit_global_atomic_float(bld, BRW_AOP_FMIN, instr);
+ break;
+ case nir_intrinsic_global_atomic_fmax:
+ nir_emit_global_atomic_float(bld, BRW_AOP_FMAX, instr);
+ break;
+ case nir_intrinsic_global_atomic_fcomp_swap:
+ nir_emit_global_atomic_float(bld, BRW_AOP_FCMPWR, instr);
+ break;
- /* Writemask */
- unsigned writemask = instr->const_index[0];
+ case nir_intrinsic_load_ssbo: {
+ assert(devinfo->gen >= 7);
- /* get_nir_src() retypes to integer. Be wary of 64-bit types though
- * since the untyped writes below operate in units of 32-bits, which
- * means that we need to write twice as many components each time.
- * Also, we have to suffle 64-bit data to be in the appropriate layout
- * expected by our 32-bit write messages.
- */
- unsigned bit_size = nir_src_bit_size(instr->src[0]);
- unsigned type_size = bit_size / 8;
+ const unsigned bit_size = nir_dest_bit_size(instr->dest);
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] =
+ get_nir_ssbo_intrinsic_index(bld, instr);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[1]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(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 num_components
- * of the block of enabled bits.
- */
- while (writemask) {
- unsigned first_component = ffs(writemask) - 1;
- unsigned num_components = ffs(~(writemask >> first_component)) - 1;
- fs_reg write_src = offset(val_reg, bld, first_component);
+ /* Make dest unsigned because that's what the temporary will be */
+ dest.type = brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_UD);
- nir_const_value *const_offset = nir_src_as_const_value(instr->src[2]);
+ /* Read the vector */
+ if (nir_intrinsic_align(instr) >= 4) {
+ assert(nir_dest_bit_size(instr->dest) == 32);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(instr->num_components);
+ fs_inst *inst =
+ bld.emit(SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
+ inst->size_written = instr->num_components * dispatch_width * 4;
+ } else {
+ assert(nir_dest_bit_size(instr->dest) <= 32);
+ assert(nir_dest_num_components(instr->dest) == 1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(bit_size);
- if (type_size > 4) {
- /* We can't write more than 2 64-bit components at once. Limit
- * the num_components of the write to what we can do and let the next
- * iteration handle the rest.
- */
- num_components = MIN2(2, num_components);
- write_src = shuffle_64bit_data_for_32bit_write(bld, write_src,
- num_components);
- } else if (type_size < 4) {
- assert(type_size == 2);
- /* For 16-bit types we pack two consecutive values into a 32-bit
- * word and use an untyped write message. For single values or not
- * 32-bit-aligned we need to use byte-scattered writes because
- * untyped writes works with 32-bit components with 32-bit
- * alignment. byte_scattered_write messages only support one
- * 16-bit component at a time. As VK_KHR_relaxed_block_layout
- * could be enabled we can not guarantee that not constant offsets
- * to be 32-bit aligned for 16-bit types. For example an array, of
- * 16-bit vec3 with array element stride of 6.
- *
- * In the case of 32-bit aligned constant offsets if there is
- * a 3-components vector we submit one untyped-write message
- * of 32-bit (first two components), and one byte-scattered
- * write message (the last component).
- */
+ fs_reg read_result = bld.vgrf(BRW_REGISTER_TYPE_UD);
+ bld.emit(SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL,
+ read_result, srcs, SURFACE_LOGICAL_NUM_SRCS);
+ bld.MOV(dest, read_result);
+ }
+ break;
+ }
- if ( !const_offset || ((const_offset->u32[0] +
- type_size * first_component) % 4)) {
- /* If we use a .yz writemask we also need to emit 2
- * byte-scattered write messages because of y-component not
- * being aligned to 32-bit.
- */
- num_components = 1;
- } else if (num_components > 2 && (num_components % 2)) {
- /* If there is an odd number of consecutive components we left
- * the not paired component for a following emit of length == 1
- * with byte_scattered_write.
- */
- num_components --;
- }
- /* For num_components == 1 we are also shuffling the component
- * because byte scattered writes of 16-bit need values to be dword
- * aligned. Shuffling only one component would be the same as
- * striding it.
- */
- write_src = shuffle_for_32bit_write(bld, write_src, 0,
- num_components);
- }
+ case nir_intrinsic_store_ssbo: {
+ assert(devinfo->gen >= 7);
- fs_reg offset_reg;
+ if (stage == MESA_SHADER_FRAGMENT)
+ brw_wm_prog_data(prog_data)->has_side_effects = true;
- if (const_offset) {
- offset_reg = brw_imm_ud(const_offset->u32[0] +
- type_size * 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),
- brw_imm_ud(type_size * first_component));
- }
+ const unsigned bit_size = nir_src_bit_size(instr->src[0]);
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] =
+ get_nir_ssbo_intrinsic_index(bld, instr);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[2]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+
+ fs_reg data = get_nir_src(instr->src[0]);
+ data.type = brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_UD);
+
+ assert(nir_intrinsic_write_mask(instr) ==
+ (1u << instr->num_components) - 1);
+ if (nir_intrinsic_align(instr) >= 4) {
+ assert(nir_src_bit_size(instr->src[0]) == 32);
+ assert(nir_src_num_components(instr->src[0]) <= 4);
+ srcs[SURFACE_LOGICAL_SRC_DATA] = data;
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(instr->num_components);
+ bld.emit(SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL,
+ fs_reg(), srcs, SURFACE_LOGICAL_NUM_SRCS);
+ } else {
+ assert(nir_src_bit_size(instr->src[0]) <= 32);
+ assert(nir_src_num_components(instr->src[0]) == 1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(bit_size);
- if (type_size < 4 && num_components == 1) {
- assert(type_size == 2);
- /* Untyped Surface messages have a fixed 32-bit size, so we need
- * to rely on byte scattered in order to write 16-bit elements.
- * The byte_scattered_write message needs that every written 16-bit
- * type to be aligned 32-bits (stride=2).
- */
- emit_byte_scattered_write(bld, surf_index, offset_reg,
- write_src,
- 1 /* dims */, 1,
- bit_size,
- BRW_PREDICATE_NONE);
- } else {
- assert(num_components * type_size <= 16);
- assert((num_components * type_size) % 4 == 0);
- assert(offset_reg.file != BRW_IMMEDIATE_VALUE ||
- offset_reg.ud % 4 == 0);
- unsigned num_slots = (num_components * type_size) / 4;
-
- emit_untyped_write(bld, surf_index, offset_reg,
- write_src,
- 1 /* dims */, num_slots,
- BRW_PREDICATE_NONE);
- }
+ srcs[SURFACE_LOGICAL_SRC_DATA] = bld.vgrf(BRW_REGISTER_TYPE_UD);
+ bld.MOV(srcs[SURFACE_LOGICAL_SRC_DATA], data);
- /* Clear the bits in the writemask that we just wrote, then try
- * again to see if more channels are left.
- */
- writemask &= (15 << (first_component + num_components));
+ bld.emit(SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL,
+ fs_reg(), srcs, SURFACE_LOGICAL_NUM_SRCS);
}
break;
}
case nir_intrinsic_store_output: {
fs_reg src = get_nir_src(instr->src[0]);
- nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
- assert(const_offset && "Indirect output stores not allowed");
-
+ unsigned store_offset = nir_src_as_uint(instr->src[1]);
unsigned num_components = instr->num_components;
unsigned first_component = nir_intrinsic_component(instr);
if (nir_src_bit_size(instr->src[0]) == 64) {
- src = shuffle_64bit_data_for_32bit_write(bld, src, num_components);
+ src = shuffle_for_32bit_write(bld, src, 0, num_components);
num_components *= 2;
}
fs_reg new_dest = retype(offset(outputs[instr->const_index[0]], bld,
- 4 * const_offset->u32[0]), src.type);
+ 4 * store_offset), src.type);
for (unsigned j = 0; j < num_components; j++) {
bld.MOV(offset(new_dest, bld, j + first_component),
offset(src, bld, j));
}
case nir_intrinsic_ssbo_atomic_add:
- nir_emit_ssbo_atomic(bld, BRW_AOP_ADD, instr);
+ nir_emit_ssbo_atomic(bld, get_op_for_atomic_add(instr, 2), instr);
break;
case nir_intrinsic_ssbo_atomic_imin:
nir_emit_ssbo_atomic(bld, BRW_AOP_IMIN, instr);
case nir_intrinsic_ssbo_atomic_comp_swap:
nir_emit_ssbo_atomic(bld, BRW_AOP_CMPWR, instr);
break;
+ case nir_intrinsic_ssbo_atomic_fmin:
+ nir_emit_ssbo_atomic_float(bld, BRW_AOP_FMIN, instr);
+ break;
+ case nir_intrinsic_ssbo_atomic_fmax:
+ nir_emit_ssbo_atomic_float(bld, BRW_AOP_FMAX, instr);
+ break;
+ case nir_intrinsic_ssbo_atomic_fcomp_swap:
+ nir_emit_ssbo_atomic_float(bld, BRW_AOP_FCMPWR, instr);
+ 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->u32[0] : 0;
+ unsigned ssbo_index = nir_src_is_const(instr->src[0]) ?
+ nir_src_as_uint(instr->src[0]) : 0;
/* A resinfo's sampler message is used to get the buffer size. The
* SIMD8's writeback message consists of four registers and SIMD16's
ubld.ADD(buffer_size, size_aligned4, negate(size_padding));
bld.MOV(retype(dest, ret_payload.type), component(buffer_size, 0));
-
- brw_mark_surface_used(prog_data, index);
break;
}
case nir_intrinsic_quad_broadcast: {
const fs_reg value = get_nir_src(instr->src[0]);
- nir_const_value *index = nir_src_as_const_value(instr->src[1]);
- assert(nir_src_bit_size(instr->src[1]) == 32);
+ const unsigned index = nir_src_as_uint(instr->src[1]);
bld.emit(SHADER_OPCODE_CLUSTER_BROADCAST, retype(dest, value.type),
- value, brw_imm_ud(index->u32[0]), brw_imm_ud(4));
+ value, brw_imm_ud(index), brw_imm_ud(4));
break;
}
const fs_reg tmp_left = horiz_stride(tmp, 2);
const fs_reg tmp_right = horiz_stride(horiz_offset(tmp, 1), 2);
- /* From the Cherryview PRM Vol. 7, "Register Region Restrictiosn":
- *
- * "When source or destination datatype is 64b or operation is
- * integer DWord multiply, regioning in Align1 must follow
- * these rules:
- *
- * [...]
- *
- * 3. Source and Destination offset must be the same, except
- * the case of scalar source."
- *
- * In order to work around this, we have to emit two 32-bit MOVs instead
- * of a single 64-bit MOV to do the shuffle.
- */
- if (type_sz(value.type) > 4 &&
- (devinfo->is_cherryview || gen_device_info_is_9lp(devinfo))) {
- ubld.MOV(subscript(tmp_left, BRW_REGISTER_TYPE_D, 0),
- subscript(src_right, BRW_REGISTER_TYPE_D, 0));
- ubld.MOV(subscript(tmp_left, BRW_REGISTER_TYPE_D, 1),
- subscript(src_right, BRW_REGISTER_TYPE_D, 1));
- ubld.MOV(subscript(tmp_right, BRW_REGISTER_TYPE_D, 0),
- subscript(src_left, BRW_REGISTER_TYPE_D, 0));
- ubld.MOV(subscript(tmp_right, BRW_REGISTER_TYPE_D, 1),
- subscript(src_left, BRW_REGISTER_TYPE_D, 1));
- } else {
- ubld.MOV(tmp_left, src_right);
- ubld.MOV(tmp_right, src_left);
- }
+ ubld.MOV(tmp_left, src_right);
+ ubld.MOV(tmp_right, src_left);
+
bld.MOV(retype(dest, value.type), tmp);
break;
}
if (nir_intrinsic_infos[instr->intrinsic].has_dest)
dest = get_nir_dest(instr->dest);
- fs_reg surface;
- 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->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]),
- brw_imm_ud(stage_prog_data->binding_table.ssbo_start));
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] = get_nir_ssbo_intrinsic_index(bld, instr);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[1]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(op);
- /* Assume this may touch any SSBO. This is the same we do for other
- * UBO/SSBO accesses with non-constant surface.
- */
- brw_mark_surface_used(prog_data,
- stage_prog_data->binding_table.ssbo_start +
- nir->info.num_ssbos - 1);
+ fs_reg data;
+ if (op != BRW_AOP_INC && op != BRW_AOP_DEC && op != BRW_AOP_PREDEC)
+ data = get_nir_src(instr->src[2]);
+
+ if (op == BRW_AOP_CMPWR) {
+ fs_reg tmp = bld.vgrf(data.type, 2);
+ fs_reg sources[2] = { data, get_nir_src(instr->src[3]) };
+ bld.LOAD_PAYLOAD(tmp, sources, 2, 0);
+ data = tmp;
}
+ srcs[SURFACE_LOGICAL_SRC_DATA] = data;
- fs_reg offset = get_nir_src(instr->src[1]);
- fs_reg data1 = get_nir_src(instr->src[2]);
- fs_reg data2;
- if (op == BRW_AOP_CMPWR)
- data2 = get_nir_src(instr->src[3]);
+ /* Emit the actual atomic operation */
+
+ bld.emit(SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
+}
+
+void
+fs_visitor::nir_emit_ssbo_atomic_float(const fs_builder &bld,
+ int op, nir_intrinsic_instr *instr)
+{
+ if (stage == MESA_SHADER_FRAGMENT)
+ brw_wm_prog_data(prog_data)->has_side_effects = true;
+
+ fs_reg dest;
+ if (nir_intrinsic_infos[instr->intrinsic].has_dest)
+ dest = get_nir_dest(instr->dest);
+
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] = get_nir_ssbo_intrinsic_index(bld, instr);
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = get_nir_src(instr->src[1]);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(op);
+
+ fs_reg data = get_nir_src(instr->src[2]);
+ if (op == BRW_AOP_FCMPWR) {
+ fs_reg tmp = bld.vgrf(data.type, 2);
+ fs_reg sources[2] = { data, get_nir_src(instr->src[3]) };
+ bld.LOAD_PAYLOAD(tmp, sources, 2, 0);
+ data = tmp;
+ }
+ srcs[SURFACE_LOGICAL_SRC_DATA] = data;
/* Emit the actual atomic 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);
+ bld.emit(SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
}
void
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;
- fs_reg data1 = get_nir_src(instr->src[1]);
- fs_reg data2;
- if (op == BRW_AOP_CMPWR)
- data2 = get_nir_src(instr->src[2]);
+ fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] = brw_imm_ud(GEN7_BTI_SLM);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(op);
+
+ fs_reg data;
+ if (op != BRW_AOP_INC && op != BRW_AOP_DEC && op != BRW_AOP_PREDEC)
+ data = get_nir_src(instr->src[1]);
+ if (op == BRW_AOP_CMPWR) {
+ fs_reg tmp = bld.vgrf(data.type, 2);
+ fs_reg sources[2] = { data, get_nir_src(instr->src[2]) };
+ bld.LOAD_PAYLOAD(tmp, sources, 2, 0);
+ data = tmp;
+ }
+ srcs[SURFACE_LOGICAL_SRC_DATA] = data;
/* Get the offset */
- nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
- if (const_offset) {
- offset = brw_imm_ud(instr->const_index[0] + const_offset->u32[0]);
+ if (nir_src_is_const(instr->src[0])) {
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] =
+ brw_imm_ud(instr->const_index[0] + nir_src_as_uint(instr->src[0]));
} else {
- offset = vgrf(glsl_type::uint_type);
- bld.ADD(offset,
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = vgrf(glsl_type::uint_type);
+ bld.ADD(srcs[SURFACE_LOGICAL_SRC_ADDRESS],
retype(get_nir_src(instr->src[0]), BRW_REGISTER_TYPE_UD),
brw_imm_ud(instr->const_index[0]));
}
/* 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);
+ bld.emit(SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
+}
+
+void
+fs_visitor::nir_emit_shared_atomic_float(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 srcs[SURFACE_LOGICAL_NUM_SRCS];
+ srcs[SURFACE_LOGICAL_SRC_SURFACE] = brw_imm_ud(GEN7_BTI_SLM);
+ srcs[SURFACE_LOGICAL_SRC_IMM_DIMS] = brw_imm_ud(1);
+ srcs[SURFACE_LOGICAL_SRC_IMM_ARG] = brw_imm_ud(op);
+
+ fs_reg data = get_nir_src(instr->src[1]);
+ if (op == BRW_AOP_FCMPWR) {
+ fs_reg tmp = bld.vgrf(data.type, 2);
+ fs_reg sources[2] = { data, get_nir_src(instr->src[2]) };
+ bld.LOAD_PAYLOAD(tmp, sources, 2, 0);
+ data = tmp;
+ }
+ srcs[SURFACE_LOGICAL_SRC_DATA] = data;
+
+ /* Get the offset */
+ if (nir_src_is_const(instr->src[0])) {
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] =
+ brw_imm_ud(instr->const_index[0] + nir_src_as_uint(instr->src[0]));
+ } else {
+ srcs[SURFACE_LOGICAL_SRC_ADDRESS] = vgrf(glsl_type::uint_type);
+ bld.ADD(srcs[SURFACE_LOGICAL_SRC_ADDRESS],
+ retype(get_nir_src(instr->src[0]), BRW_REGISTER_TYPE_UD),
+ brw_imm_ud(instr->const_index[0]));
+ }
+
+ /* Emit the actual atomic operation operation */
+
+ bld.emit(SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL,
+ dest, srcs, SURFACE_LOGICAL_NUM_SRCS);
+}
+
+void
+fs_visitor::nir_emit_global_atomic(const fs_builder &bld,
+ int op, nir_intrinsic_instr *instr)
+{
+ if (stage == MESA_SHADER_FRAGMENT)
+ brw_wm_prog_data(prog_data)->has_side_effects = true;
+
+ fs_reg dest;
+ if (nir_intrinsic_infos[instr->intrinsic].has_dest)
+ dest = get_nir_dest(instr->dest);
+
+ fs_reg addr = get_nir_src(instr->src[0]);
+
+ fs_reg data;
+ if (op != BRW_AOP_INC && op != BRW_AOP_DEC && op != BRW_AOP_PREDEC)
+ data = get_nir_src(instr->src[1]);
+
+ if (op == BRW_AOP_CMPWR) {
+ fs_reg tmp = bld.vgrf(data.type, 2);
+ fs_reg sources[2] = { data, get_nir_src(instr->src[2]) };
+ bld.LOAD_PAYLOAD(tmp, sources, 2, 0);
+ data = tmp;
+ }
+
+ bld.emit(SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL,
+ dest, addr, data, brw_imm_ud(op));
+}
+
+void
+fs_visitor::nir_emit_global_atomic_float(const fs_builder &bld,
+ int op, nir_intrinsic_instr *instr)
+{
+ if (stage == MESA_SHADER_FRAGMENT)
+ brw_wm_prog_data(prog_data)->has_side_effects = true;
+
+ assert(nir_intrinsic_infos[instr->intrinsic].has_dest);
+ fs_reg dest = get_nir_dest(instr->dest);
+
+ fs_reg addr = get_nir_src(instr->src[0]);
+
+ assert(op != BRW_AOP_INC && op != BRW_AOP_DEC && op != BRW_AOP_PREDEC);
+ fs_reg data = get_nir_src(instr->src[1]);
+
+ if (op == BRW_AOP_FCMPWR) {
+ fs_reg tmp = bld.vgrf(data.type, 2);
+ fs_reg sources[2] = { data, get_nir_src(instr->src[2]) };
+ bld.LOAD_PAYLOAD(tmp, sources, 2, 0);
+ data = tmp;
+ }
+
+ bld.emit(SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL,
+ dest, addr, data, brw_imm_ud(op));
}
void
break;
}
break;
+ case nir_tex_src_min_lod:
+ srcs[TEX_LOGICAL_SRC_MIN_LOD] =
+ retype(get_nir_src_imm(instr->src[i].src), BRW_REGISTER_TYPE_F);
+ break;
case nir_tex_src_ms_index:
srcs[TEX_LOGICAL_SRC_SAMPLE_INDEX] = retype(src, BRW_REGISTER_TYPE_UD);
break;
case nir_tex_src_offset: {
nir_const_value *const_offset =
nir_src_as_const_value(instr->src[i].src);
+ assert(nir_src_bit_size(instr->src[i].src) == 32);
unsigned offset_bits = 0;
if (const_offset &&
brw_texture_offset(const_offset->i32,
unreachable("should be lowered");
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 {
- max_used += stage_prog_data->binding_table.texture_start;
- }
- brw_mark_surface_used(prog_data, max_used);
-
/* Emit code to evaluate the actual indexing expression */
fs_reg tmp = vgrf(glsl_type::uint_type);
bld.ADD(tmp, src, brw_imm_ud(texture));
break;
case nir_tex_src_plane: {
- nir_const_value *const_plane =
- nir_src_as_const_value(instr->src[i].src);
- const uint32_t plane = const_plane->u32[0];
+ const uint32_t plane = nir_src_as_uint(instr->src[i].src);
const uint32_t texture_index =
instr->texture_index +
stage_prog_data->binding_table.plane_start[plane] -
}
}
-/**
- * This helper takes the result of a load operation that reads 32-bit elements
- * in this format:
- *
- * x x x x x x x x
- * y y y y y y y y
- * z z z z z z z z
- * w w w w w w w w
- *
- * and shuffles the data to get this:
- *
- * x y x y x y x y
- * x y x y x y x y
- * z w z w z w z w
- * z w z w z w z w
- *
- * Which is exactly what we want if the load is reading 64-bit components
- * like doubles, where x represents the low 32-bit of the x double component
- * and y represents the high 32-bit of the x double component (likewise with
- * z and w for double component y). The parameter @components represents
- * the number of 64-bit components present in @src. This would typically be
- * 2 at most, since we can only fit 2 double elements in the result of a
- * vec4 load.
- *
- * Notice that @dst and @src can be the same register.
- */
-void
-shuffle_32bit_load_result_to_64bit_data(const fs_builder &bld,
- const fs_reg &dst,
- const fs_reg &src,
- uint32_t components)
-{
- assert(type_sz(src.type) == 4);
- assert(type_sz(dst.type) == 8);
-
- /* A temporary that we will use to shuffle the 32-bit data of each
- * component in the vector into valid 64-bit data. We can't write directly
- * to dst because dst can be (and would usually be) the same as src
- * and in that case the first MOV in the loop below would overwrite the
- * data read in the second MOV.
- */
- fs_reg tmp = bld.vgrf(dst.type);
-
- for (unsigned i = 0; i < components; i++) {
- const fs_reg component_i = offset(src, bld, 2 * i);
-
- bld.MOV(subscript(tmp, src.type, 0), component_i);
- bld.MOV(subscript(tmp, src.type, 1), offset(component_i, bld, 1));
-
- bld.MOV(offset(dst, bld, i), tmp);
- }
-}
-
-/**
- * This helper does the inverse operation of
- * SHUFFLE_32BIT_LOAD_RESULT_TO_64BIT_DATA.
- *
- * We need to do this when we are going to use untyped write messsages that
- * operate with 32-bit components in order to arrange our 64-bit data to be
- * in the expected layout.
- *
- * Notice that callers of this function, unlike in the case of the inverse
- * operation, would typically need to call this with dst and src being
- * different registers, since they would otherwise corrupt the original
- * 64-bit data they are about to write. Because of this the function checks
- * that the src and dst regions involved in the operation do not overlap.
- */
-fs_reg
-shuffle_64bit_data_for_32bit_write(const fs_builder &bld,
- const fs_reg &src,
- uint32_t components)
-{
- assert(type_sz(src.type) == 8);
-
- fs_reg dst = bld.vgrf(BRW_REGISTER_TYPE_D, 2 * components);
-
- for (unsigned i = 0; i < components; i++) {
- const fs_reg component_i = offset(src, bld, i);
- bld.MOV(offset(dst, bld, 2 * i), subscript(component_i, dst.type, 0));
- bld.MOV(offset(dst, bld, 2 * i + 1), subscript(component_i, dst.type, 1));
- }
-
- return dst;
-}
-
/*
* This helper takes a source register and un/shuffles it into the destination
* register.
return component(retype(tmp, BRW_REGISTER_TYPE_DF), 0);
}
+
+fs_reg
+setup_imm_b(const fs_builder &bld, int8_t v)
+{
+ const fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_B);
+ bld.MOV(tmp, brw_imm_w(v));
+ return tmp;
+}
+
+fs_reg
+setup_imm_ub(const fs_builder &bld, uint8_t v)
+{
+ const fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UB);
+ bld.MOV(tmp, brw_imm_uw(v));
+ return tmp;
+}