}
}
-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());
- unsigned num_iter = 1;
- if (type->is_dual_slot())
- num_iter = 2;
- for (unsigned count = 0; count < num_iter; count++) {
- this->outputs[*location] = *reg;
- *reg = offset(*reg, bld, 4);
- (*location)++;
- }
- }
-}
-
void
fs_visitor::nir_setup_outputs()
{
if (stage == MESA_SHADER_TESS_CTRL || stage == MESA_SHADER_FRAGMENT)
return;
- nir_outputs = bld.vgrf(BRW_REGISTER_TYPE_F, nir->num_outputs);
-
nir_foreach_variable(var, &nir->outputs) {
- switch (stage) {
- case MESA_SHADER_VERTEX:
- case MESA_SHADER_TESS_EVAL:
- case MESA_SHADER_GEOMETRY: {
- fs_reg reg = offset(nir_outputs, bld, var->data.driver_location);
- unsigned location = var->data.location;
- nir_setup_single_output_varying(®, var->type, &location);
- break;
- }
- default:
- unreachable("unhandled shader stage");
+ const unsigned vec4s =
+ var->data.compact ? DIV_ROUND_UP(glsl_get_length(var->type), 4)
+ : type_size_vec4(var->type);
+ fs_reg reg = bld.vgrf(BRW_REGISTER_TYPE_F, 4 * vec4s);
+ for (unsigned i = 0; i < vec4s; i++) {
+ if (outputs[var->data.driver_location + i].file == BAD_FILE)
+ outputs[var->data.driver_location + i] = offset(reg, bld, 4 * i);
}
}
}
g0.negate = true;
}
- tmp.type = BRW_REGISTER_TYPE_W;
- tmp.subreg_offset = 2;
- tmp.stride = 2;
-
- bld.OR(tmp, g0, brw_imm_uw(0x3f80));
-
- tmp.type = BRW_REGISTER_TYPE_D;
- tmp.subreg_offset = 0;
- tmp.stride = 1;
+ bld.OR(subscript(tmp, BRW_REGISTER_TYPE_W, 1),
+ g0, brw_imm_uw(0x3f80));
} else {
/* Bit 31 of g1.6 is 0 if the polygon is front facing. */
fs_reg g1_6 = fs_reg(retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_D));
case nir_op_f2d:
case nir_op_i2d:
case nir_op_u2d:
+ case nir_op_f2i64:
+ case nir_op_f2u64:
+ case nir_op_i2i64:
+ case nir_op_i2u64:
+ case nir_op_u2i64:
+ case nir_op_u2u64:
+ case nir_op_b2i64:
/* CHV PRM, vol07, 3D Media GPGPU Engine, Register Region Restrictions:
*
* "When source or destination is 64b (...), regioning in Align1
* data elements aligned to 64-bit. This restriction does not apply to
* BDW and later.
*/
- if (devinfo->is_cherryview || devinfo->is_broxton) {
+ if (nir_dest_bit_size(instr->dest.dest) == 64 &&
+ nir_src_bit_size(instr->src[0].src) == 32 &&
+ (devinfo->is_cherryview || devinfo->is_broxton)) {
fs_reg tmp = bld.vgrf(result.type, 1);
tmp = subscript(tmp, op[0].type, 0);
inst = bld.MOV(tmp, op[0]);
case nir_op_d2f:
case nir_op_d2i:
case nir_op_d2u:
- inst = bld.MOV(result, op[0]);
- inst->saturate = instr->dest.saturate;
+ if (instr->op == nir_op_b2i64) {
+ bld.MOV(result, negate(op[0]));
+ } else {
+ inst = bld.MOV(result, op[0]);
+ inst->saturate = instr->dest.saturate;
+ }
break;
case nir_op_f2i:
break;
}
+ case nir_op_pack_int_2x32_split:
+ bld.emit(FS_OPCODE_PACK, result, op[0], op[1]);
+ break;
+
+ case nir_op_unpack_int_2x32_split_x:
+ case nir_op_unpack_int_2x32_split_y: {
+ if (instr->op == nir_op_unpack_int_2x32_split_x)
+ bld.MOV(result, subscript(op[0], BRW_REGISTER_TYPE_UD, 0));
+ else
+ bld.MOV(result, subscript(op[0], BRW_REGISTER_TYPE_UD, 1));
+ break;
+ }
+
case nir_op_fpow:
inst = bld.emit(SHADER_OPCODE_POW, result, op[0], op[1]);
inst->saturate = instr->dest.saturate;
glsl_interp_mode interpolation)
{
struct brw_wm_prog_data *wm_prog_data =
- (struct brw_wm_prog_data *) bld.shader->stage_prog_data;
+ brw_wm_prog_data(bld.shader->stage_prog_data);
fs_inst *inst;
fs_reg payload;
int mlen;
inst = bld.emit(opcode, dst, payload, desc);
inst->mlen = mlen;
/* 2 floats per slot returned */
- inst->regs_written = 2 * bld.dispatch_width() / 8;
+ inst->size_written = 2 * dst.component_size(inst->exec_size);
inst->pi_noperspective = interpolation == INTERP_MODE_NOPERSPECTIVE;
wm_prog_data->pulls_bary = true;
{
assert(stage == MESA_SHADER_GEOMETRY);
- struct brw_gs_prog_data *gs_prog_data =
- (struct brw_gs_prog_data *) prog_data;
+ struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);
if (gs_compile->control_data_header_size_bits == 0)
return;
assert(stage == MESA_SHADER_GEOMETRY);
assert(gs_compile->control_data_bits_per_vertex != 0);
- struct brw_gs_prog_data *gs_prog_data =
- (struct brw_gs_prog_data *) prog_data;
+ struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);
const fs_builder abld = bld.annotate("emit control data bits");
const fs_builder fwa_bld = bld.exec_all();
{
assert(stage == MESA_SHADER_GEOMETRY);
- struct brw_gs_prog_data *gs_prog_data =
- (struct brw_gs_prog_data *) prog_data;
+ struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);
fs_reg vertex_count = get_nir_src(vertex_count_nir_src);
vertex_count.type = BRW_REGISTER_TYPE_UD;
* be recorded by transform feedback, we can simply discard all geometry
* bound to these streams when transform feedback is disabled.
*/
- if (stream_id > 0 && !nir->info.has_transform_feedback_varyings)
+ if (stream_id > 0 && !nir->info->has_transform_feedback_varyings)
return;
/* If we're outputting 32 control data bits or less, then we can wait
unsigned num_components,
unsigned first_component)
{
- struct brw_gs_prog_data *gs_prog_data = (struct brw_gs_prog_data *) prog_data;
+ 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);
} else {
for (unsigned i = 0; i < num_components; i++) {
bld.MOV(offset(dst, bld, i),
- fs_reg(ATTR, imm_offset + i, dst.type));
+ fs_reg(ATTR, imm_offset + i + first_component, dst.type));
}
}
return;
/* 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.
+ * 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));
+ brw_imm_ud(nir->info->gs.vertices_in * REG_SIZE));
}
} else {
assert(gs_prog_data->invocations > 1);
/* Use first_icp_handle as the base offset. There is one DWord
* of URB handles per vertex, so inform the register allocator that
- * we might read up to ceil(nir->info.gs.vertices_in / 8) registers.
+ * we might read up to ceil(nir->info->gs.vertices_in / 8) 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(DIV_ROUND_UP(nir->info.gs.vertices_in, 8) *
+ brw_imm_ud(DIV_ROUND_UP(nir->info->gs.vertices_in, 8) *
REG_SIZE));
}
}
unsigned read_components = num_components + first_component;
fs_reg tmp = bld.vgrf(dst.type, read_components);
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp, icp_handle);
- inst->regs_written = read_components * type_sz(tmp_dst.type) / 4;
+ inst->size_written = read_components *
+ tmp.component_size(inst->exec_size);
for (unsigned i = 0; i < num_components; i++) {
bld.MOV(offset(tmp_dst, bld, i),
offset(tmp, bld, i + first_component));
} else {
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp_dst,
icp_handle);
- inst->regs_written = num_components * type_sz(tmp_dst.type) / 4;
+ inst->size_written = num_components *
+ tmp_dst.component_size(inst->exec_size);
}
inst->offset = base_offset + offset_const->u32[0];
inst->mlen = 1;
if (first_component != 0) {
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp,
payload);
- inst->regs_written = read_components * type_sz(tmp_dst.type) / 4;
+ inst->size_written = read_components *
+ tmp.component_size(inst->exec_size);
for (unsigned i = 0; i < num_components; i++) {
bld.MOV(offset(tmp_dst, bld, i),
offset(tmp, bld, i + first_component));
} else {
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp_dst,
payload);
- inst->regs_written = num_components * type_sz(tmp_dst.type) / 4;
+ inst->size_written = num_components *
+ tmp_dst.component_size(inst->exec_size);
}
inst->offset = base_offset;
inst->mlen = 2;
/* 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;
+ inst->size_written = 4 * REG_SIZE;
bld.MOV(dst, offset(tmp, bld, 3));
}
}
{
assert(stage == MESA_SHADER_TESS_CTRL);
struct brw_tcs_prog_key *tcs_key = (struct brw_tcs_prog_key *) key;
- struct brw_tcs_prog_data *tcs_prog_data =
- (struct brw_tcs_prog_data *) prog_data;
+ struct brw_tcs_prog_data *tcs_prog_data = brw_tcs_prog_data(prog_data);
fs_reg dst;
if (nir_intrinsic_infos[instr->intrinsic].has_dest)
inst->offset = imm_offset;
inst->mlen = 2;
}
- inst->regs_written =
- ((num_components + first_component) * type_sz(dst.type) / 4);
+ inst->size_written = (num_components + first_component) *
+ inst->dst.component_size(inst->exec_size);
/* If we are reading 64-bit data using 32-bit read messages we need
* build proper 64-bit data elements by shuffling the low and high
if (inst->offset == 0 && indirect_offset.file == BAD_FILE) {
assert(type_sz(dst.type) < 8);
inst->dst = bld.vgrf(dst.type, 4);
- inst->regs_written = 4;
+ inst->size_written = 4 * REG_SIZE;
bld.MOV(dst, offset(inst->dst, bld, 3));
}
bld.MOV(patch_handle,
retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD));
- if (imm_offset == 0) {
- /* This is a read of gl_TessLevelInner[], which lives in the
- * Patch URB header. The layout depends on the domain.
- */
- dst.type = BRW_REGISTER_TYPE_F;
- switch (tcs_key->tes_primitive_mode) {
- case GL_QUADS: {
- /* DWords 3-2 (reversed) */
- fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_F, 4);
-
- inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp, patch_handle);
- inst->offset = 0;
- inst->mlen = 1;
- inst->regs_written = 4;
-
- /* dst.xy = tmp.wz */
- bld.MOV(dst, offset(tmp, bld, 3));
- bld.MOV(offset(dst, bld, 1), offset(tmp, bld, 2));
- break;
- }
- case GL_TRIANGLES:
- /* DWord 4; hardcode offset = 1 and regs_written = 1 */
- inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dst, patch_handle);
- inst->offset = 1;
- inst->mlen = 1;
- inst->regs_written = 1;
- break;
- case GL_ISOLINES:
- /* All channels are undefined. */
- break;
- default:
- unreachable("Bogus tessellation domain");
- }
- } else if (imm_offset == 1) {
- /* This is a read of gl_TessLevelOuter[], which lives in the
- * Patch URB header. The layout depends on the domain.
- */
- dst.type = BRW_REGISTER_TYPE_F;
-
- fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_F, 4);
- inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp, patch_handle);
- inst->offset = 1;
- inst->mlen = 1;
- inst->regs_written = 4;
-
- /* Reswizzle: WZYX */
- fs_reg srcs[4] = {
- offset(tmp, bld, 3),
- offset(tmp, bld, 2),
- offset(tmp, bld, 1),
- offset(tmp, bld, 0),
- };
-
- unsigned num_components;
- switch (tcs_key->tes_primitive_mode) {
- case GL_QUADS:
- num_components = 4;
- break;
- case GL_TRIANGLES:
- num_components = 3;
- break;
- case GL_ISOLINES:
- /* Isolines are not reversed; swizzle .zw -> .xy */
- srcs[0] = offset(tmp, bld, 2);
- srcs[1] = offset(tmp, bld, 3);
- num_components = 2;
- break;
- default:
- unreachable("Bogus tessellation domain");
- }
- bld.LOAD_PAYLOAD(dst, srcs, num_components, 0);
- } else {
+ {
if (first_component != 0) {
unsigned read_components =
instr->num_components + first_component;
fs_reg tmp = bld.vgrf(dst.type, read_components);
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp,
patch_handle);
- inst->regs_written = read_components;
+ inst->size_written = read_components * REG_SIZE;
for (unsigned i = 0; i < instr->num_components; i++) {
bld.MOV(offset(dst, bld, i),
offset(tmp, bld, i + first_component));
} else {
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dst,
patch_handle);
- inst->regs_written = instr->num_components;
+ inst->size_written = instr->num_components * REG_SIZE;
}
inst->offset = imm_offset;
inst->mlen = 1;
fs_reg tmp = bld.vgrf(dst.type, read_components);
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp,
payload);
- inst->regs_written = read_components;
+ inst->size_written = read_components * REG_SIZE;
for (unsigned i = 0; i < instr->num_components; i++) {
bld.MOV(offset(dst, bld, i),
offset(tmp, bld, i + first_component));
} else {
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, dst,
payload);
- inst->regs_written = instr->num_components;
+ inst->size_written = instr->num_components * REG_SIZE;
}
inst->offset = imm_offset;
inst->mlen = 2;
if (indirect_offset.file != BAD_FILE) {
srcs[header_regs++] = indirect_offset;
- } else if (!is_passthrough_shader) {
- if (imm_offset == 0) {
- value.type = BRW_REGISTER_TYPE_F;
-
- mask &= (1 << tesslevel_inner_components(tcs_key->tes_primitive_mode)) - 1;
-
- /* This is a write to gl_TessLevelInner[], which lives in the
- * Patch URB header. The layout depends on the domain.
- */
- switch (tcs_key->tes_primitive_mode) {
- case GL_QUADS:
- /* gl_TessLevelInner[].xy lives at DWords 3-2 (reversed).
- * We use an XXYX swizzle to reverse put .xy in the .wz
- * channels, and use a .zw writemask.
- */
- mask = writemask_for_backwards_vector(mask);
- swiz = BRW_SWIZZLE4(0, 0, 1, 0);
- break;
- case GL_TRIANGLES:
- /* gl_TessLevelInner[].x lives at DWord 4, so we set the
- * writemask to X and bump the URB offset by 1.
- */
- imm_offset = 1;
- break;
- case GL_ISOLINES:
- /* Skip; gl_TessLevelInner[] doesn't exist for isolines. */
- return;
- default:
- unreachable("Bogus tessellation domain");
- }
- } else if (imm_offset == 1) {
- /* This is a write to gl_TessLevelOuter[] which lives in the
- * Patch URB Header at DWords 4-7. However, it's reversed, so
- * instead of .xyzw we have .wzyx.
- */
- value.type = BRW_REGISTER_TYPE_F;
-
- mask &= (1 << tesslevel_outer_components(tcs_key->tes_primitive_mode)) - 1;
-
- if (tcs_key->tes_primitive_mode == GL_ISOLINES) {
- /* Isolines .xy should be stored in .zw, in order. */
- swiz = BRW_SWIZZLE4(0, 0, 0, 1);
- mask <<= 2;
- } else {
- /* Other domains are reversed; store .wzyx instead of .xyzw */
- swiz = BRW_SWIZZLE_WZYX;
- mask = writemask_for_backwards_vector(mask);
- }
- }
}
if (mask == 0)
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;
+ struct brw_tes_prog_data *tes_prog_data = brw_tes_prog_data(prog_data);
fs_reg dest;
if (nir_intrinsic_infos[instr->intrinsic].has_dest)
}
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), 6 + 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);
fs_reg tmp = bld.vgrf(dest.type, read_components);
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp,
patch_handle);
- inst->regs_written = read_components;
+ inst->size_written = read_components * REG_SIZE;
for (unsigned i = 0; i < instr->num_components; i++) {
bld.MOV(offset(dest, bld, i),
offset(tmp, bld, i + first_component));
} else {
inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dest,
patch_handle);
- inst->regs_written = instr->num_components;
+ inst->size_written = instr->num_components * REG_SIZE;
}
inst->mlen = 1;
inst->offset = imm_offset;
}
inst->mlen = 2;
inst->offset = imm_offset;
- inst->regs_written =
- ((num_components + first_component) * type_sz(dest.type) / 4);
+ inst->size_written = (num_components + first_component) *
+ inst->dst.component_size(inst->exec_size);
/* If we are reading 64-bit data using 32-bit read messages we need
* build proper 64-bit data elements by shuffling the low and high
switch (instr->intrinsic) {
case nir_intrinsic_load_primitive_id:
assert(stage == MESA_SHADER_GEOMETRY);
- assert(((struct brw_gs_prog_data *)prog_data)->include_primitive_id);
+ assert(brw_gs_prog_data(prog_data)->include_primitive_id);
bld.MOV(retype(dest, BRW_REGISTER_TYPE_UD),
retype(fs_reg(brw_vec8_grf(2, 0)), BRW_REGISTER_TYPE_UD));
break;
const brw_wm_prog_key *wm_key =
reinterpret_cast<const brw_wm_prog_key *>(key);
assert(!wm_key->coherent_fb_fetch);
- const brw_wm_prog_data *wm_prog_data =
- reinterpret_cast<const brw_wm_prog_data *>(stage_prog_data);
+ const struct brw_wm_prog_data *wm_prog_data =
+ brw_wm_prog_data(stage_prog_data);
/* Calculate the surface index relative to the start of the texture binding
* table block, since that's what the texturing messages expect.
STATIC_ASSERT(ARRAY_SIZE(srcs) == TEX_LOGICAL_NUM_SRCS);
fs_inst *inst = bld.emit(op, dst, srcs, ARRAY_SIZE(srcs));
- inst->regs_written = 4 * inst->dst.component_size(inst->exec_size) /
- REG_SIZE;
+ inst->size_written = 4 * inst->dst.component_size(inst->exec_size);
return inst;
}
assert(bld.shader->devinfo->gen >= 9);
fs_inst *inst = bld.emit(FS_OPCODE_FB_READ_LOGICAL, dst);
inst->target = target;
- inst->regs_written = 4 * inst->dst.component_size(inst->exec_size) /
- REG_SIZE;
+ inst->size_written = 4 * inst->dst.component_size(inst->exec_size);
return inst;
}
break;
}
+ case nir_intrinsic_load_layer_id:
+ dest.type = BRW_REGISTER_TYPE_UD;
+ bld.MOV(dest, fetch_render_target_array_index(bld));
+ break;
+
case nir_intrinsic_load_helper_invocation:
case nir_intrinsic_load_sample_mask_in:
case nir_intrinsic_load_sample_id: {
nir_intrinsic_instr *instr)
{
assert(stage == MESA_SHADER_COMPUTE);
- struct brw_cs_prog_data *cs_prog_data =
- (struct brw_cs_prog_data *) prog_data;
+ struct brw_cs_prog_data *cs_prog_data = brw_cs_prog_data(prog_data);
fs_reg dest;
if (nir_intrinsic_infos[instr->intrinsic].has_dest)
switch (instr->intrinsic) {
case nir_intrinsic_atomic_counter_inc:
case nir_intrinsic_atomic_counter_dec:
- case nir_intrinsic_atomic_counter_read: {
+ case nir_intrinsic_atomic_counter_read:
+ case nir_intrinsic_atomic_counter_add:
+ case nir_intrinsic_atomic_counter_min:
+ case nir_intrinsic_atomic_counter_max:
+ case nir_intrinsic_atomic_counter_and:
+ case nir_intrinsic_atomic_counter_or:
+ case nir_intrinsic_atomic_counter_xor:
+ case nir_intrinsic_atomic_counter_exchange:
+ case nir_intrinsic_atomic_counter_comp_swap: {
if (stage == MESA_SHADER_FRAGMENT &&
instr->intrinsic != nir_intrinsic_atomic_counter_read)
- ((struct brw_wm_prog_data *)prog_data)->has_side_effects = true;
+ brw_wm_prog_data(prog_data)->has_side_effects = true;
+
+ /* Get some metadata from the image intrinsic. */
+ const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic];
/* 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 +
instr->const_index[0]);
+ const fs_reg src0 = (info->num_srcs >= 2
+ ? get_nir_src(instr->src[1]) : fs_reg());
+ const fs_reg src1 = (info->num_srcs >= 3
+ ? get_nir_src(instr->src[2]) : fs_reg());
fs_reg tmp;
+ assert(info->num_srcs <= 3);
+
/* Emit a surface read or atomic op. */
- switch (instr->intrinsic) {
- case nir_intrinsic_atomic_counter_read:
+ if (instr->intrinsic == nir_intrinsic_atomic_counter_read) {
tmp = emit_untyped_read(bld, brw_imm_ud(surface), offset, 1, 1);
- break;
-
- case nir_intrinsic_atomic_counter_inc:
- 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, brw_imm_ud(surface), offset, fs_reg(),
- fs_reg(), 1, 1, BRW_AOP_PREDEC);
- break;
-
- default:
- unreachable("Unreachable");
+ } else {
+ tmp = emit_untyped_atomic(bld, brw_imm_ud(surface), offset, src0,
+ src1, 1, 1,
+ get_atomic_counter_op(instr->intrinsic));
}
/* Assign the result. */
if (stage == MESA_SHADER_FRAGMENT &&
instr->intrinsic != nir_intrinsic_image_load)
- ((struct brw_wm_prog_data *)prog_data)->has_side_effects = true;
+ 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 fs_builder ubld = bld.group(8, 0);
const fs_reg tmp = ubld.vgrf(BRW_REGISTER_TYPE_UD, 2);
ubld.emit(SHADER_OPCODE_MEMORY_FENCE, tmp)
- ->regs_written = 2;
+ ->size_written = 2 * REG_SIZE;
break;
}
case nir_intrinsic_shader_clock: {
/* We cannot do anything if there is an event, so ignore it for now */
- fs_reg shader_clock = get_timestamp(bld);
- const fs_reg srcs[] = { shader_clock.set_smear(0), shader_clock.set_smear(1) };
-
+ const fs_reg shader_clock = get_timestamp(bld);
+ const fs_reg srcs[] = { component(shader_clock, 0),
+ component(shader_clock, 1) };
bld.LOAD_PAYLOAD(dest, srcs, ARRAY_SIZE(srcs), 0);
break;
}
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;
+ src.offset = const_offset->u32[0];
for (unsigned j = 0; j < instr->num_components; j++) {
bld.MOV(offset(dest, bld, j), offset(src, bld, j));
*/
brw_mark_surface_used(prog_data,
stage_prog_data->binding_table.ubo_start +
- nir->info.num_ubos - 1);
+ nir->info->num_ubos - 1);
}
nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
* and we have to split it if necessary.
*/
const unsigned type_size = type_sz(dest.type);
- const fs_reg packed_consts = bld.vgrf(BRW_REGISTER_TYPE_F);
+ const unsigned block_sz = 64; /* Fetch one cacheline at a time. */
+ const fs_builder ubld = bld.exec_all().group(block_sz / 4, 0);
+ 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;
-
- /* Number of usable components in the next 16B-aligned load */
+ /* Number of usable components in the next block-aligned load. */
const unsigned count = MIN2(instr->num_components - c,
- (16 - base % 16) / type_size);
+ (block_sz - base % block_sz) / type_size);
- bld.exec_all()
- .emit(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
- packed_consts, surf_index, brw_imm_ud(base & ~15));
+ ubld.emit(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
+ packed_consts, surf_index,
+ brw_imm_ud(base & ~(block_sz - 1)));
const fs_reg consts =
- retype(byte_offset(packed_consts, base & 15), dest.type);
+ retype(byte_offset(packed_consts, base & (block_sz - 1)),
+ dest.type);
for (unsigned d = 0; d < count; d++)
bld.MOV(offset(dest, bld, c + d), component(consts, d));
*/
brw_mark_surface_used(prog_data,
stage_prog_data->binding_table.ssbo_start +
- nir->info.num_ssbos - 1);
+ nir->info->num_ssbos - 1);
}
fs_reg offset_reg;
assert(devinfo->gen >= 7);
if (stage == MESA_SHADER_FRAGMENT)
- ((struct brw_wm_prog_data *)prog_data)->has_side_effects = true;
+ brw_wm_prog_data(prog_data)->has_side_effects = true;
/* Block index */
fs_reg surf_index;
brw_mark_surface_used(prog_data,
stage_prog_data->binding_table.ssbo_start +
- nir->info.num_ssbos - 1);
+ nir->info->num_ssbos - 1);
}
/* Value */
case nir_intrinsic_store_output: {
fs_reg src = get_nir_src(instr->src[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]);
+ fs_reg new_dest = retype(offset(outputs[instr->const_index[0]], bld,
+ 4 * const_offset->u32[0]), src.type);
unsigned num_components = instr->num_components;
unsigned first_component = nir_intrinsic_component(instr);
src_payload, brw_imm_ud(index));
inst->header_size = 0;
inst->mlen = 1;
- inst->regs_written = 4;
+ inst->size_written = 4 * REG_SIZE;
bld.MOV(retype(dest, ret_payload.type), component(ret_payload, 0));
brw_mark_surface_used(prog_data, index);
int op, nir_intrinsic_instr *instr)
{
if (stage == MESA_SHADER_FRAGMENT)
- ((struct brw_wm_prog_data *)prog_data)->has_side_effects = true;
+ brw_wm_prog_data(prog_data)->has_side_effects = true;
fs_reg dest;
if (nir_intrinsic_infos[instr->intrinsic].has_dest)
*/
brw_mark_surface_used(prog_data,
stage_prog_data->binding_table.ssbo_start +
- nir->info.num_ssbos - 1);
+ nir->info->num_ssbos - 1);
}
fs_reg offset = get_nir_src(instr->src[1]);
if (instr->sampler_dim == GLSL_SAMPLER_DIM_BUF)
srcs[TEX_LOGICAL_SRC_LOD] = brw_imm_d(0);
+ uint32_t header_bits = 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) {
srcs[TEX_LOGICAL_SRC_LOD] =
retype(get_nir_src_imm(instr->src[i].src), BRW_REGISTER_TYPE_F);
break;
- case nir_tex_src_comparitor:
+ case nir_tex_src_comparator:
srcs[TEX_LOGICAL_SRC_SHADOW_C] = retype(src, BRW_REGISTER_TYPE_F);
break;
case nir_tex_src_coord:
case nir_tex_src_offset: {
nir_const_value *const_offset =
nir_src_as_const_value(instr->src[i].src);
- if (const_offset) {
- unsigned header_bits = brw_texture_offset(const_offset->i32, 3);
- if (header_bits != 0)
- srcs[TEX_LOGICAL_SRC_OFFSET_VALUE] = brw_imm_ud(header_bits);
+ unsigned offset_bits = 0;
+ if (const_offset &&
+ brw_texture_offset(const_offset->i32,
+ nir_tex_instr_src_size(instr, i),
+ &offset_bits)) {
+ header_bits |= offset_bits;
} else {
- srcs[TEX_LOGICAL_SRC_OFFSET_VALUE] =
+ srcs[TEX_LOGICAL_SRC_TG4_OFFSET] =
retype(src, BRW_REGISTER_TYPE_D);
}
break;
opcode = SHADER_OPCODE_LOD_LOGICAL;
break;
case nir_texop_tg4:
- if (srcs[TEX_LOGICAL_SRC_OFFSET_VALUE].file != BAD_FILE &&
- srcs[TEX_LOGICAL_SRC_OFFSET_VALUE].file != IMM)
+ if (srcs[TEX_LOGICAL_SRC_TG4_OFFSET].file != BAD_FILE)
opcode = SHADER_OPCODE_TG4_OFFSET_LOGICAL;
else
opcode = SHADER_OPCODE_TG4_LOGICAL;
unreachable("unknown texture opcode");
}
+ if (instr->op == nir_texop_tg4) {
+ if (instr->component == 1 &&
+ key_tex->gather_channel_quirk_mask & (1 << texture)) {
+ /* gather4 sampler is broken for green channel on RG32F --
+ * we must ask for blue instead.
+ */
+ header_bits |= 2 << 16;
+ } else {
+ header_bits |= instr->component << 16;
+ }
+ }
+
fs_reg dst = bld.vgrf(brw_type_for_nir_type(instr->dest_type), 4);
fs_inst *inst = bld.emit(opcode, dst, srcs, ARRAY_SIZE(srcs));
+ inst->offset = header_bits;
const unsigned dest_size = nir_tex_instr_dest_size(instr);
if (devinfo->gen >= 9 &&
nir_ssa_def_components_read(&instr->dest.ssa):
(1 << dest_size) - 1;
assert(write_mask != 0); /* dead code should have been eliminated */
- inst->regs_written = util_last_bit(write_mask) * dispatch_width / 8;
+ inst->size_written = util_last_bit(write_mask) *
+ inst->dst.component_size(inst->exec_size);
} else {
- inst->regs_written = 4 * dispatch_width / 8;
+ inst->size_written = 4 * inst->dst.component_size(inst->exec_size);
}
if (srcs[TEX_LOGICAL_SRC_SHADOW_C].file != BAD_FILE)
inst->shadow_compare = true;
- if (srcs[TEX_LOGICAL_SRC_OFFSET_VALUE].file == IMM)
- inst->offset = srcs[TEX_LOGICAL_SRC_OFFSET_VALUE].ud;
-
- if (instr->op == nir_texop_tg4) {
- if (instr->component == 1 &&
- key_tex->gather_channel_quirk_mask & (1 << texture)) {
- /* gather4 sampler is broken for green channel on RG32F --
- * we must ask for blue instead.
- */
- inst->offset |= 2 << 16;
- } else {
- inst->offset |= instr->component << 16;
- }
-
- if (devinfo->gen == 6)
- emit_gen6_gather_wa(key_tex->gen6_gather_wa[texture], dst);
- }
+ if (instr->op == nir_texop_tg4 && devinfo->gen == 6)
+ emit_gen6_gather_wa(key_tex->gen6_gather_wa[texture], dst);
fs_reg nir_dest[4];
for (unsigned i = 0; i < dest_size; i++)
assert(type_sz(src.type) == 8);
assert(type_sz(dst.type) == 4);
- assert(!src.in_range(dst, 2 * components * bld.dispatch_width() / 8));
+ assert(!regions_overlap(
+ dst, 2 * components * dst.component_size(bld.dispatch_width()),
+ src, components * src.component_size(bld.dispatch_width())));
for (unsigned i = 0; i < components; i++) {
const fs_reg component_i = offset(src, bld, i);
* instruction allows to set the 64-bit immediate value.
*/
if (devinfo->is_haswell) {
- const fs_builder ubld = bld.exec_all();
+ const fs_builder ubld = bld.exec_all().group(1, 0);
fs_reg dst = ubld.vgrf(BRW_REGISTER_TYPE_DF, 1);
ubld.DIM(dst, brw_imm_df(v));
return component(dst, 0);
projects / mesa.git / blobdiff