continue;
if (devinfo->gen >= 7) {
- /* The offset arg is a vec4-aligned immediate byte offset. */
- fs_reg const_offset_reg = inst->src[1];
- assert(const_offset_reg.file == IMM &&
- const_offset_reg.type == BRW_REGISTER_TYPE_UD);
- assert(const_offset_reg.ud % 16 == 0);
-
- fs_reg payload, offset;
- if (devinfo->gen >= 9) {
- /* We have to use a message header on Skylake to get SIMD4x2
- * mode. Reserve space for the register.
- */
- offset = payload = fs_reg(VGRF, alloc.allocate(2));
- offset.offset += REG_SIZE;
- inst->mlen = 2;
- } else {
- offset = payload = fs_reg(VGRF, alloc.allocate(1));
- inst->mlen = 1;
- }
-
- /* This is actually going to be a MOV, but since only the first dword
- * is accessed, we have a special opcode to do just that one. Note
- * that this needs to be an operation that will be considered a def
- * by live variable analysis, or register allocation will explode.
- */
- fs_inst *setup = new(mem_ctx) fs_inst(FS_OPCODE_SET_SIMD4X2_OFFSET,
- 8, offset, const_offset_reg);
- setup->force_writemask_all = true;
+ const fs_builder ubld = fs_builder(this, block, inst).exec_all();
+ const fs_reg payload = ubld.group(8, 0).vgrf(BRW_REGISTER_TYPE_UD);
- setup->ir = inst->ir;
- setup->annotation = inst->annotation;
- inst->insert_before(block, setup);
+ ubld.group(8, 0).MOV(payload,
+ retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
+ ubld.group(1, 0).MOV(component(payload, 2),
+ brw_imm_ud(inst->src[1].ud / 16));
- /* Similarly, this will only populate the first 4 channels of the
- * result register (since we only use smear values from 0-3), but we
- * don't tell the optimizer.
- */
inst->opcode = FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7;
inst->src[1] = payload;
+ inst->header_size = 1;
+ inst->mlen = 1;
invalidate_live_intervals();
} else {
fs_generator::generate_uniform_pull_constant_load_gen7(fs_inst *inst,
struct brw_reg dst,
struct brw_reg index,
- struct brw_reg offset)
+ struct brw_reg payload)
{
assert(index.type == BRW_REGISTER_TYPE_UD);
-
- assert(offset.file == BRW_GENERAL_REGISTER_FILE);
- /* Reference just the dword we need, to avoid angering validate_reg(). */
- offset = brw_vec1_grf(offset.nr, 0);
-
- /* We use the SIMD4x2 mode because we want to end up with 4 components in
- * the destination loaded consecutively from the same offset (which appears
- * in the first component, and the rest are ignored).
- */
- dst.width = BRW_WIDTH_4;
-
- struct brw_reg src = offset;
- bool header_present = false;
-
- if (devinfo->gen >= 9) {
- /* Skylake requires a message header in order to use SIMD4x2 mode. */
- src = retype(brw_vec4_grf(offset.nr, 0), BRW_REGISTER_TYPE_UD);
- header_present = true;
-
- brw_push_insn_state(p);
- brw_set_default_mask_control(p, BRW_MASK_DISABLE);
- brw_set_default_exec_size(p, BRW_EXECUTE_8);
- brw_MOV(p, vec8(src), retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
- brw_set_default_access_mode(p, BRW_ALIGN_1);
-
- brw_MOV(p, get_element_ud(src, 2),
- brw_imm_ud(GEN9_SAMPLER_SIMD_MODE_EXTENSION_SIMD4X2));
- brw_pop_insn_state(p);
- }
+ assert(payload.file == BRW_GENERAL_REGISTER_FILE);
if (index.file == BRW_IMMEDIATE_VALUE) {
-
- uint32_t surf_index = index.ud;
+ const uint32_t surf_index = index.ud;
brw_push_insn_state(p);
brw_set_default_compression_control(p, BRW_COMPRESSION_NONE);
brw_inst_set_exec_size(devinfo, send, BRW_EXECUTE_4);
brw_pop_insn_state(p);
- brw_set_dest(p, send, retype(dst, BRW_REGISTER_TYPE_UD));
- brw_set_src0(p, send, src);
- brw_set_sampler_message(p, send,
+ brw_set_dest(p, send, vec4(retype(dst, BRW_REGISTER_TYPE_UD)));
+ brw_set_src0(p, send, vec4(retype(payload, BRW_REGISTER_TYPE_UD)));
+ brw_set_dp_read_message(p, send,
surf_index,
- 0, /* LD message ignores sampler unit */
- GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
- 1, /* rlen */
- inst->mlen,
- header_present,
- BRW_SAMPLER_SIMD_MODE_SIMD4X2,
- 0);
- } else {
+ BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW,
+ GEN7_DATAPORT_DC_OWORD_BLOCK_READ,
+ GEN6_SFID_DATAPORT_CONSTANT_CACHE,
+ 1, /* mlen */
+ true, /* header */
+ 1); /* rlen */
+ } else {
struct brw_reg addr = vec1(retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD));
brw_push_insn_state(p);
/* dst = send(payload, a0.0 | <descriptor>) */
brw_inst *insn = brw_send_indirect_message(
- p, BRW_SFID_SAMPLER, dst, src, addr);
- brw_set_sampler_message(p, insn,
- 0,
- 0, /* LD message ignores sampler unit */
- GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
- 1, /* rlen */
- inst->mlen,
- header_present,
- BRW_SAMPLER_SIMD_MODE_SIMD4X2,
- 0);
+ p, GEN6_SFID_DATAPORT_CONSTANT_CACHE,
+ vec4(retype(dst, BRW_REGISTER_TYPE_UD)),
+ vec4(retype(payload, BRW_REGISTER_TYPE_UD)), addr);
+ brw_inst_set_exec_size(p->devinfo, insn, BRW_EXECUTE_4);
+ brw_set_dp_read_message(p, insn,
+ 0, /* surface */
+ BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW,
+ GEN7_DATAPORT_DC_OWORD_BLOCK_READ,
+ GEN6_SFID_DATAPORT_CONSTANT_CACHE,
+ 1, /* mlen */
+ true, /* header */
+ 1); /* rlen */
brw_pop_insn_state(p);
}
projects / mesa.git / commitdiff