brw_set_dest(p, insn, brw_null_reg());
brw_set_src0(p, insn, payload);
- brw_set_src1(p, insn, brw_imm_d(0));
+ brw_set_src1(p, insn, brw_imm_ud(0u));
brw_inst_set_sfid(p->devinfo, insn, BRW_SFID_URB);
brw_inst_set_urb_opcode(p->devinfo, insn, GEN8_URB_OPCODE_SIMD8_WRITE);
brw_set_dest(p, insn, retype(brw_null_reg(), BRW_REGISTER_TYPE_UW));
brw_set_src0(p, insn, retype(payload, BRW_REGISTER_TYPE_UW));
- brw_set_src1(p, insn, brw_imm_d(0));
+ brw_set_src1(p, insn, brw_imm_ud(0u));
/* Terminate a compute shader by sending a message to the thread spawner.
*/
return true;
} else if (devinfo->has_pln) {
- /* From the Sandy Bridge PRM Vol. 4, Pt. 2, Section 8.3.53, "Plane":
- *
- * "[DevSNB]:<src1> must be even register aligned.
- *
- * This restriction is lifted on Ivy Bridge.
- */
- assert(devinfo->gen >= 7 || (delta_x.nr & 1) == 0);
- brw_PLN(p, dst, interp, delta_x);
+ if (devinfo->gen <= 6 && (delta_x.nr & 1) != 0) {
+ /* From the Sandy Bridge PRM Vol. 4, Pt. 2, Section 8.3.53, "Plane":
+ *
+ * "[DevSNB]:<src1> must be even register aligned.
+ *
+ * This restriction is lifted on Ivy Bridge.
+ *
+ * This means that we need to split PLN into LINE+MAC on-the-fly.
+ * Unfortunately, the inputs are laid out for PLN and not LINE+MAC so
+ * we have to split into SIMD8 pieces. For gen4 (!has_pln), the
+ * coordinate registers are laid out differently so we leave it as a
+ * SIMD16 instruction.
+ */
+ assert(inst->exec_size == 8 || inst->exec_size == 16);
+ assert(inst->group % 16 == 0);
- return false;
+ brw_push_insn_state(p);
+ brw_set_default_exec_size(p, BRW_EXECUTE_8);
+
+ /* Thanks to two accumulators, we can emit all the LINEs and then all
+ * the MACs. This improves parallelism a bit.
+ */
+ for (unsigned g = 0; g < inst->exec_size / 8; g++) {
+ brw_inst *line = brw_LINE(p, brw_null_reg(), interp,
+ offset(delta_x, g * 2));
+ brw_inst_set_group(devinfo, line, inst->group + g * 8);
+
+ /* LINE writes the accumulator automatically on gen4-5. On Sandy
+ * Bridge and later, we have to explicitly enable it.
+ */
+ if (devinfo->gen >= 6)
+ brw_inst_set_acc_wr_control(p->devinfo, line, true);
+
+ /* brw_set_default_saturate() is called before emitting
+ * instructions, so the saturate bit is set in each instruction,
+ * so we need to unset it on the LINE instructions.
+ */
+ brw_inst_set_saturate(p->devinfo, line, false);
+ }
+
+ for (unsigned g = 0; g < inst->exec_size / 8; g++) {
+ brw_inst *mac = brw_MAC(p, offset(dst, g), suboffset(interp, 1),
+ offset(delta_x, g * 2 + 1));
+ brw_inst_set_group(devinfo, mac, inst->group + g * 8);
+ brw_inst_set_cond_modifier(p->devinfo, mac, inst->conditional_mod);
+ }
+
+ brw_pop_insn_state(p);
+
+ return true;
+ } else {
+ brw_PLN(p, dst, interp, delta_x);
+
+ return false;
+ }
} else {
i[0] = brw_LINE(p, brw_null_reg(), interp, delta_x);
i[1] = brw_MAC(p, dst, suboffset(interp, 1), delta_y);
uint32_t return_format;
bool is_combined_send = inst->eot;
+ /* Sampler EOT message of less than the dispatch width would kill the
+ * thread prematurely.
+ */
+ assert(!is_combined_send || inst->exec_size == dispatch_width);
+
switch (dst.type) {
case BRW_REGISTER_TYPE_D:
return_format = BRW_SAMPLER_RETURN_FORMAT_SINT32;
}
break;
case SHADER_OPCODE_TXS:
+ case SHADER_OPCODE_IMAGE_SIZE:
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_RESINFO;
break;
case SHADER_OPCODE_TXD:
}
}
- uint32_t base_binding_table_index = (inst->opcode == SHADER_OPCODE_TG4 ||
- inst->opcode == SHADER_OPCODE_TG4_OFFSET)
- ? prog_data->binding_table.gather_texture_start
- : prog_data->binding_table.texture_start;
+ uint32_t base_binding_table_index;
+ switch (inst->opcode) {
+ case SHADER_OPCODE_TG4:
+ case SHADER_OPCODE_TG4_OFFSET:
+ base_binding_table_index = prog_data->binding_table.gather_texture_start;
+ break;
+ case SHADER_OPCODE_IMAGE_SIZE:
+ base_binding_table_index = prog_data->binding_table.image_start;
+ break;
+ default:
+ base_binding_table_index = prog_data->binding_table.texture_start;
+ break;
+ }
if (surface_index.file == BRW_IMMEDIATE_VALUE &&
sampler_index.file == BRW_IMMEDIATE_VALUE) {
brw_pop_insn_state(p);
/* dst = send(offset, a0.0 | <descriptor>) */
- brw_inst *insn = brw_send_indirect_message(
- p, BRW_SFID_SAMPLER, dst, src, addr);
- brw_set_sampler_message(p, insn,
- 0 /* surface */,
- 0 /* sampler */,
- msg_type,
- inst->size_written / REG_SIZE,
- inst->mlen /* mlen */,
- inst->header_size != 0 /* header */,
- simd_mode,
- return_format);
+ brw_send_indirect_message(
+ p, BRW_SFID_SAMPLER, dst, src, addr,
+ brw_message_desc(devinfo, inst->mlen, inst->size_written / REG_SIZE,
+ inst->header_size) |
+ brw_sampler_desc(devinfo,
+ 0 /* surface */,
+ 0 /* sampler */,
+ msg_type,
+ simd_mode,
+ return_format));
/* visitor knows more than we do about the surface limit required,
* so has already done marking.
brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
brw_pop_insn_state(p);
+ brw_inst_set_sfid(devinfo, send, GEN6_SFID_DATAPORT_CONSTANT_CACHE);
brw_set_dest(p, send, retype(dst, BRW_REGISTER_TYPE_UD));
brw_set_src0(p, send, retype(payload, BRW_REGISTER_TYPE_UD));
- brw_set_dp_read_message(p, send, surf_index,
- BRW_DATAPORT_OWORD_BLOCK_DWORDS(inst->exec_size),
- GEN7_DATAPORT_DC_OWORD_BLOCK_READ,
- GEN6_SFID_DATAPORT_CONSTANT_CACHE,
- 1, /* mlen */
- true, /* header */
- DIV_ROUND_UP(inst->size_written, REG_SIZE));
+ brw_set_desc(p, send,
+ brw_message_desc(devinfo, 1, DIV_ROUND_UP(inst->size_written,
+ REG_SIZE), true) |
+ brw_dp_read_desc(devinfo, surf_index,
+ BRW_DATAPORT_OWORD_BLOCK_DWORDS(inst->exec_size),
+ GEN7_DATAPORT_DC_OWORD_BLOCK_READ,
+ BRW_DATAPORT_READ_TARGET_DATA_CACHE));
} else {
struct brw_reg addr = vec1(retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD));
brw_set_src1(p, insn_and, brw_imm_ud(0x0ff));
/* dst = send(payload, a0.0 | <descriptor>) */
- brw_inst *insn = brw_send_indirect_message(
+ brw_send_indirect_message(
p, GEN6_SFID_DATAPORT_CONSTANT_CACHE,
retype(dst, BRW_REGISTER_TYPE_UD),
- retype(payload, BRW_REGISTER_TYPE_UD), addr);
- brw_set_dp_read_message(p, insn, 0 /* surface */,
- BRW_DATAPORT_OWORD_BLOCK_DWORDS(inst->exec_size),
- GEN7_DATAPORT_DC_OWORD_BLOCK_READ,
- GEN6_SFID_DATAPORT_CONSTANT_CACHE,
- 1, /* mlen */
- true, /* header */
- DIV_ROUND_UP(inst->size_written, REG_SIZE));
+ retype(payload, BRW_REGISTER_TYPE_UD), addr,
+ brw_message_desc(devinfo, 1,
+ DIV_ROUND_UP(inst->size_written, REG_SIZE), true) |
+ brw_dp_read_desc(devinfo, 0 /* surface */,
+ BRW_DATAPORT_OWORD_BLOCK_DWORDS(inst->exec_size),
+ GEN7_DATAPORT_DC_OWORD_BLOCK_READ,
+ BRW_DATAPORT_READ_TARGET_DATA_CACHE));
brw_pop_insn_state(p);
}
brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
brw_inst_set_compression(devinfo, send, false);
+ brw_inst_set_sfid(devinfo, send, BRW_SFID_SAMPLER);
brw_set_dest(p, send, retype(dst, BRW_REGISTER_TYPE_UW));
brw_set_src0(p, send, header);
if (devinfo->gen < 6)
* stored in it.
*/
uint32_t return_format = BRW_SAMPLER_RETURN_FORMAT_FLOAT32;
- brw_set_sampler_message(p, send,
- surf_index,
- 0, /* sampler (unused) */
- msg_type,
- rlen,
- inst->mlen,
- inst->header_size != 0,
- simd_mode,
- return_format);
+ brw_set_desc(p, send,
+ brw_message_desc(devinfo, inst->mlen, rlen, inst->header_size) |
+ brw_sampler_desc(devinfo, surf_index,
+ 0, /* sampler (unused) */
+ msg_type, simd_mode, return_format));
}
void
* gen7, so the fact that it's a send message is hidden at the IR level.
*/
assert(inst->header_size == 0);
- assert(!inst->mlen);
+ assert(inst->mlen);
assert(index.type == BRW_REGISTER_TYPE_UD);
- uint32_t simd_mode, rlen, mlen;
+ uint32_t simd_mode, rlen;
if (inst->exec_size == 16) {
- mlen = 2;
rlen = 8;
simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD16;
} else {
assert(inst->exec_size == 8);
- mlen = 1;
rlen = 4;
simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD8;
}
uint32_t surf_index = index.ud;
brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
+ brw_inst_set_sfid(devinfo, send, BRW_SFID_SAMPLER);
brw_set_dest(p, send, retype(dst, BRW_REGISTER_TYPE_UW));
brw_set_src0(p, send, offset);
- brw_set_sampler_message(p, send,
- surf_index,
- 0, /* LD message ignores sampler unit */
- GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
- rlen,
- mlen,
- false, /* no header */
- simd_mode,
- 0);
+ brw_set_desc(p, send,
+ brw_message_desc(devinfo, inst->mlen, rlen, false) |
+ brw_sampler_desc(devinfo, surf_index,
+ 0, /* LD message ignores sampler unit */
+ GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
+ simd_mode, 0));
} else {
brw_pop_insn_state(p);
/* dst = send(offset, a0.0 | <descriptor>) */
- brw_inst *insn = brw_send_indirect_message(
+ brw_send_indirect_message(
p, BRW_SFID_SAMPLER, retype(dst, BRW_REGISTER_TYPE_UW),
- offset, addr);
- brw_set_sampler_message(p, insn,
- 0 /* surface */,
- 0 /* sampler */,
- GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
- rlen /* rlen */,
- mlen /* mlen */,
- false /* header */,
- simd_mode,
- 0);
+ offset, addr,
+ brw_message_desc(devinfo, inst->mlen, rlen, false) |
+ brw_sampler_desc(devinfo,
+ 0 /* surface */,
+ 0 /* sampler */,
+ GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
+ simd_mode,
+ 0));
}
}
-/**
- * Cause the current pixel/sample mask (from R1.7 bits 15:0) to be transferred
- * into the flags register (f0.0).
- *
- * Used only on Gen6 and above.
- */
-void
-fs_generator::generate_mov_dispatch_to_flags(fs_inst *inst)
-{
- struct brw_reg flags = brw_flag_subreg(inst->flag_subreg);
- struct brw_reg dispatch_mask;
-
- if (devinfo->gen >= 6)
- dispatch_mask = retype(brw_vec1_grf(1, 7), BRW_REGISTER_TYPE_UW);
- else
- dispatch_mask = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
-
- brw_push_insn_state(p);
- brw_set_default_mask_control(p, BRW_MASK_DISABLE);
- brw_set_default_exec_size(p, BRW_EXECUTE_1);
- brw_MOV(p, flags, dispatch_mask);
- brw_pop_insn_state(p);
-}
-
void
fs_generator::generate_pixel_interpolator_query(fs_inst *inst,
struct brw_reg dst,
struct brw_reg msg_data,
unsigned msg_type)
{
- assert(inst->size_written % REG_SIZE == 0);
+ const bool has_payload = inst->src[0].file != BAD_FILE;
assert(msg_data.type == BRW_REGISTER_TYPE_UD);
+ assert(inst->size_written % REG_SIZE == 0);
brw_pixel_interpolator_query(p,
retype(dst, BRW_REGISTER_TYPE_UW),
- src,
+ /* If we don't have a payload, what we send doesn't matter */
+ has_payload ? src : brw_vec8_grf(0, 0),
inst->pi_noperspective,
msg_type,
msg_data,
- inst->mlen,
+ has_payload ? 2 * inst->exec_size / 8 : 1,
inst->size_written / REG_SIZE);
}
case SHADER_OPCODE_SAMPLEINFO:
generate_tex(inst, dst, src[0], src[1], src[2]);
break;
+
+ case SHADER_OPCODE_IMAGE_SIZE:
+ generate_tex(inst, dst, src[0], src[1], brw_imm_ud(0));
+ break;
+
case FS_OPCODE_DDX_COARSE:
case FS_OPCODE_DDX_FINE:
generate_ddx(inst, dst, src[0]);
generate_fb_read(inst, dst, src[0]);
break;
- case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
- generate_mov_dispatch_to_flags(inst);
- break;
-
case FS_OPCODE_DISCARD_JUMP:
generate_discard_jump(inst);
break;
inst->header_size);
break;
+ case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT:
+ assert(src[2].file == BRW_IMMEDIATE_VALUE);
+ brw_untyped_atomic_float(p, dst, src[0], src[1], src[2].ud,
+ inst->mlen, !inst->dst.is_null(),
+ inst->header_size);
+ break;
+
case SHADER_OPCODE_UNTYPED_SURFACE_READ:
assert(!inst->header_size);
assert(src[2].file == BRW_IMMEDIATE_VALUE);
projects / mesa.git / blobdiff