}
}
+void
+fs_generator::generate_quad_swizzle(const fs_inst *inst,
+ struct brw_reg dst, struct brw_reg src,
+ unsigned swiz)
+{
+ /* Requires a quad. */
+ assert(inst->exec_size >= 4);
+
+ if (src.file == BRW_IMMEDIATE_VALUE ||
+ has_scalar_region(src)) {
+ /* The value is uniform across all channels */
+ brw_MOV(p, dst, src);
+
+ } else if (devinfo->gen < 11 && type_sz(src.type) == 4) {
+ /* This only works on 8-wide 32-bit values */
+ assert(inst->exec_size == 8);
+ assert(src.hstride == BRW_HORIZONTAL_STRIDE_1);
+ assert(src.vstride == src.width + 1);
+ brw_set_default_access_mode(p, BRW_ALIGN_16);
+ struct brw_reg swiz_src = stride(src, 4, 4, 1);
+ swiz_src.swizzle = swiz;
+ brw_MOV(p, dst, swiz_src);
+
+ } else {
+ assert(src.hstride == BRW_HORIZONTAL_STRIDE_1);
+ assert(src.vstride == src.width + 1);
+ const struct brw_reg src_0 = suboffset(src, BRW_GET_SWZ(swiz, 0));
+
+ switch (swiz) {
+ case BRW_SWIZZLE_XXXX:
+ case BRW_SWIZZLE_YYYY:
+ case BRW_SWIZZLE_ZZZZ:
+ case BRW_SWIZZLE_WWWW:
+ brw_MOV(p, dst, stride(src_0, 4, 4, 0));
+ break;
+
+ case BRW_SWIZZLE_XXZZ:
+ case BRW_SWIZZLE_YYWW:
+ brw_MOV(p, dst, stride(src_0, 2, 2, 0));
+ break;
+
+ case BRW_SWIZZLE_XYXY:
+ case BRW_SWIZZLE_ZWZW:
+ assert(inst->exec_size == 4);
+ brw_MOV(p, dst, stride(src_0, 0, 2, 1));
+ break;
+
+ default:
+ assert(inst->force_writemask_all);
+ brw_set_default_exec_size(p, cvt(inst->exec_size / 4) - 1);
+
+ for (unsigned c = 0; c < 4; c++) {
+ brw_inst *insn = brw_MOV(
+ p, stride(suboffset(dst, c),
+ 4 * inst->dst.stride, 1, 4 * inst->dst.stride),
+ stride(suboffset(src, BRW_GET_SWZ(swiz, c)), 4, 1, 0));
+
+ brw_inst_set_no_dd_clear(devinfo, insn, c < 3);
+ brw_inst_set_no_dd_check(devinfo, insn, c > 0);
+ }
+
+ break;
+ }
+ }
+}
+
void
fs_generator::generate_urb_read(fs_inst *inst,
struct brw_reg dst,
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.
*/
}
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));
}
}
brw_F32TO16(p, dst_w, x);
}
-void
-fs_generator::generate_unpack_half_2x16_split(fs_inst *inst,
- struct brw_reg dst,
- struct brw_reg src)
-{
- assert(devinfo->gen >= 7);
- assert(dst.type == BRW_REGISTER_TYPE_F);
- assert(src.type == BRW_REGISTER_TYPE_UD);
-
- /* From the Ivybridge PRM, Vol4, Part3, Section 6.26 f16to32:
- *
- * Because this instruction does not have a 16-bit floating-point type,
- * the source data type must be Word (W). The destination type must be
- * F (Float).
- */
- struct brw_reg src_w = spread(retype(src, BRW_REGISTER_TYPE_W), 2);
-
- /* Each channel of src has the form of unpackHalf2x16's input: 0xhhhhllll.
- * For the Y case, we wish to access only the upper word; therefore
- * a 16-bit subregister offset is needed.
- */
- assert(inst->opcode == FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X ||
- inst->opcode == FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y);
- if (inst->opcode == FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y)
- src_w.subnr += 2;
-
- brw_F16TO32(p, dst, src_w);
-}
-
void
fs_generator::generate_shader_time_add(fs_inst *,
struct brw_reg payload,
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]);
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);
break;
case SHADER_OPCODE_QUAD_SWIZZLE:
- /* This only works on 8-wide 32-bit values */
- assert(inst->exec_size == 8);
- assert(type_sz(src[0].type) == 4);
- assert(inst->force_writemask_all);
assert(src[1].file == BRW_IMMEDIATE_VALUE);
assert(src[1].type == BRW_REGISTER_TYPE_UD);
-
- if (src[0].file == BRW_IMMEDIATE_VALUE ||
- (src[0].vstride == 0 && src[0].hstride == 0)) {
- /* The value is uniform across all channels */
- brw_MOV(p, dst, src[0]);
- } else {
- brw_set_default_access_mode(p, BRW_ALIGN_16);
- struct brw_reg swiz_src = stride(src[0], 4, 4, 1);
- swiz_src.swizzle = inst->src[1].ud;
- brw_MOV(p, dst, swiz_src);
- }
+ generate_quad_swizzle(inst, dst, src[0], src[1].ud);
break;
case SHADER_OPCODE_CLUSTER_BROADCAST: {
generate_pack_half_2x16_split(inst, dst, src[0], src[1]);
break;
- case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
- case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
- generate_unpack_half_2x16_split(inst, dst, src[0]);
- break;
-
case FS_OPCODE_PLACEHOLDER_HALT:
/* This is the place where the final HALT needs to be inserted if
* we've emitted any discards. If not, this will emit no code.
projects / mesa.git / blobdiff