#include "brw_vec4.h"
#include "brw_cfg.h"
#include "brw_eu.h"
-#include "common/gen_debug.h"
+#include "dev/gen_debug.h"
using namespace brw;
inst->header_size != 0,
BRW_SAMPLER_SIMD_MODE_SIMD4X2,
return_format);
-
- brw_mark_surface_used(&prog_data->base, sampler + base_binding_table_index);
} else {
/* Non-constant sampler index. */
gen6_resolve_implied_move(p, &src, inst->base_mrf);
/* 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,
- 1 /* rlen */,
- inst->mlen /* mlen */,
- inst->header_size != 0 /* header */,
- BRW_SAMPLER_SIMD_MODE_SIMD4X2,
- return_format);
+ brw_send_indirect_message(
+ p, BRW_SFID_SAMPLER, dst, src, addr,
+ brw_message_desc(devinfo, inst->mlen, 1, inst->header_size) |
+ brw_sampler_desc(devinfo,
+ 0 /* surface */,
+ 0 /* sampler */,
+ msg_type,
+ BRW_SAMPLER_SIMD_MODE_SIMD4X2,
+ return_format),
+ false /* EOT */);
/* 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_set_dest(p, send, brw_null_reg());
brw_set_src0(p, send, urb_header);
+ brw_set_desc(p, send, brw_message_desc(devinfo, inst->mlen, 0, true));
- brw_set_message_descriptor(p, send, BRW_SFID_URB,
- inst->mlen /* mlen */, 0 /* rlen */,
- true /* header */, false /* eot */);
+ brw_inst_set_sfid(devinfo, send, BRW_SFID_URB);
brw_inst_set_urb_opcode(devinfo, send, BRW_URB_OPCODE_WRITE_OWORD);
brw_inst_set_urb_global_offset(devinfo, send, inst->offset);
if (inst->urb_write_flags & BRW_URB_WRITE_EOT) {
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, dst, header);
+
+ /* Uniforms will have a stride <0;4,1>, and we need to convert to <0;1,0>.
+ * Other values get <4;1,0>.
+ */
+ struct brw_reg restrided_offset;
+ if (offset.vstride == BRW_VERTICAL_STRIDE_0 &&
+ offset.width == BRW_WIDTH_4 &&
+ offset.hstride == BRW_HORIZONTAL_STRIDE_1) {
+ restrided_offset = stride(offset, 0, 1, 0);
+ } else {
+ restrided_offset = stride(offset, 4, 1, 0);
+ }
+
/* m0.3-0.4: 128-bit-granular offsets into the URB from the handles */
- brw_MOV(p, vec2(get_element_ud(dst, 3)), stride(offset, 4, 1, 0));
+ brw_MOV(p, vec2(get_element_ud(dst, 3)), restrided_offset);
brw_pop_insn_state(p);
}
brw_set_dest(p, send, dst);
brw_set_src0(p, send, header);
- brw_set_message_descriptor(p, send, BRW_SFID_URB,
- 1 /* mlen */, 1 /* rlen */,
- true /* header */, false /* eot */);
+ brw_set_desc(p, send, brw_message_desc(devinfo, 1, 1, true));
+
+ brw_inst_set_sfid(devinfo, send, BRW_SFID_URB);
brw_inst_set_urb_opcode(devinfo, send, BRW_URB_OPCODE_READ_OWORD);
brw_inst_set_urb_swizzle_control(devinfo, send, BRW_URB_SWIZZLE_INTERLEAVE);
brw_inst_set_urb_per_slot_offset(devinfo, send, 1);
brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
brw_set_dest(p, send, brw_null_reg());
brw_set_src0(p, send, header);
- brw_set_message_descriptor(p, send, BRW_SFID_URB,
- 1 /* mlen */, 0 /* rlen */,
- true /* header */, false /* eot */);
+ brw_set_desc(p, send, brw_message_desc(devinfo, 1, 0, true));
+
+ brw_inst_set_sfid(devinfo, send, BRW_SFID_URB);
brw_inst_set_urb_opcode(devinfo, send, BRW_URB_OPCODE_READ_OWORD);
brw_inst_set_urb_complete(devinfo, send, 1);
brw_inst_set_urb_swizzle_control(devinfo, send, is_unpaired.ud ?
const unsigned target_cache =
devinfo->gen >= 7 ? GEN7_SFID_DATAPORT_DATA_CACHE :
devinfo->gen >= 6 ? GEN6_SFID_DATAPORT_RENDER_CACHE :
- BRW_DATAPORT_READ_TARGET_RENDER_CACHE;
+ BRW_SFID_DATAPORT_READ;
/* Each of the 8 channel enables is considered for whether each
* dword is written.
*/
brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
+ brw_inst_set_sfid(devinfo, send, target_cache);
brw_set_dest(p, send, dst);
brw_set_src0(p, send, header);
if (devinfo->gen < 6)
brw_inst_set_cond_modifier(devinfo, send, inst->base_mrf);
- brw_set_dp_read_message(p, send,
- brw_scratch_surface_idx(p),
- BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
- msg_type, target_cache,
- 2, /* mlen */
- true, /* header_present */
- 1 /* rlen */);
+ brw_set_desc(p, send,
+ brw_message_desc(devinfo, 2, 1, true) |
+ brw_dp_read_desc(devinfo,
+ brw_scratch_surface_idx(p),
+ BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
+ msg_type, BRW_DATAPORT_READ_TARGET_RENDER_CACHE));
}
static void
const unsigned target_cache =
(devinfo->gen >= 7 ? GEN7_SFID_DATAPORT_DATA_CACHE :
devinfo->gen >= 6 ? GEN6_SFID_DATAPORT_RENDER_CACHE :
- BRW_DATAPORT_READ_TARGET_RENDER_CACHE);
+ BRW_SFID_DATAPORT_WRITE);
struct brw_reg header = brw_vec8_grf(0, 0);
bool write_commit;
* dword is written.
*/
brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
+ brw_inst_set_sfid(p->devinfo, send, target_cache);
brw_set_dest(p, send, dst);
brw_set_src0(p, send, header);
if (devinfo->gen < 6)
brw_inst_set_cond_modifier(p->devinfo, send, inst->base_mrf);
- brw_set_dp_write_message(p, send,
- brw_scratch_surface_idx(p),
- BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
- msg_type,
- target_cache,
- 3, /* mlen */
- true, /* header present */
- false, /* not a render target write */
- write_commit, /* rlen */
- false, /* eot */
- write_commit);
+ brw_set_desc(p, send,
+ brw_message_desc(devinfo, 3, write_commit, true) |
+ brw_dp_write_desc(devinfo,
+ brw_scratch_surface_idx(p),
+ BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
+ msg_type,
+ false, /* not a render target write */
+ write_commit));
}
static void
const struct gen_device_info *devinfo = p->devinfo;
const unsigned target_cache =
(devinfo->gen >= 6 ? GEN6_SFID_DATAPORT_SAMPLER_CACHE :
- BRW_DATAPORT_READ_TARGET_DATA_CACHE);
+ BRW_SFID_DATAPORT_READ);
assert(index.file == BRW_IMMEDIATE_VALUE &&
index.type == BRW_REGISTER_TYPE_UD);
uint32_t surf_index = index.ud;
* dword is written.
*/
brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
+ brw_inst_set_sfid(devinfo, send, target_cache);
brw_set_dest(p, send, dst);
brw_set_src0(p, send, header);
if (devinfo->gen < 6)
brw_inst_set_cond_modifier(p->devinfo, send, inst->base_mrf);
- brw_set_dp_read_message(p, send,
- surf_index,
- BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
- msg_type,
- target_cache,
- 2, /* mlen */
- true, /* header_present */
- 1 /* rlen */);
+ brw_set_desc(p, send,
+ brw_message_desc(devinfo, 2, 1, true) |
+ brw_dp_read_desc(devinfo, surf_index,
+ BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
+ msg_type,
+ BRW_DATAPORT_READ_TARGET_DATA_CACHE));
}
static void
inst->header_size > 0,
BRW_SAMPLER_SIMD_MODE_SIMD4X2,
BRW_SAMPLER_RETURN_FORMAT_SINT32);
-
- brw_mark_surface_used(&prog_data->base, surf_index.ud);
}
static void
struct brw_reg surf_index,
struct brw_reg offset)
{
+ const struct gen_device_info *devinfo = p->devinfo;
assert(surf_index.type == BRW_REGISTER_TYPE_UD);
if (surf_index.file == BRW_IMMEDIATE_VALUE) {
brw_inst *insn = brw_next_insn(p, BRW_OPCODE_SEND);
+ brw_inst_set_sfid(devinfo, insn, BRW_SFID_SAMPLER);
brw_set_dest(p, insn, dst);
brw_set_src0(p, insn, offset);
- brw_set_sampler_message(p, insn,
- surf_index.ud,
- 0, /* LD message ignores sampler unit */
- GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
- 1, /* rlen */
- inst->mlen,
- inst->header_size != 0,
- BRW_SAMPLER_SIMD_MODE_SIMD4X2,
- 0);
-
- brw_mark_surface_used(&prog_data->base, surf_index.ud);
-
+ brw_set_desc(p, insn,
+ brw_message_desc(devinfo, inst->mlen, 1, inst->header_size) |
+ brw_sampler_desc(devinfo, surf_index.ud,
+ 0, /* LD message ignores sampler unit */
+ GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
+ BRW_SAMPLER_SIMD_MODE_SIMD4X2, 0));
} else {
struct brw_reg addr = vec1(retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD));
/* a0.0 = surf_index & 0xff */
brw_inst *insn_and = brw_next_insn(p, BRW_OPCODE_AND);
- brw_inst_set_exec_size(p->devinfo, insn_and, BRW_EXECUTE_1);
+ brw_inst_set_exec_size(devinfo, insn_and, BRW_EXECUTE_1);
brw_set_dest(p, insn_and, addr);
brw_set_src0(p, insn_and, vec1(retype(surf_index, BRW_REGISTER_TYPE_UD)));
brw_set_src1(p, insn_and, brw_imm_ud(0x0ff));
brw_pop_insn_state(p);
/* dst = send(offset, a0.0 | <descriptor>) */
- brw_inst *insn = brw_send_indirect_message(
- p, BRW_SFID_SAMPLER, dst, offset, addr);
- brw_set_sampler_message(p, insn,
- 0 /* surface */,
- 0 /* sampler */,
- GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
- 1 /* rlen */,
- inst->mlen,
- inst->header_size != 0,
- BRW_SAMPLER_SIMD_MODE_SIMD4X2,
- 0);
+ brw_send_indirect_message(
+ p, BRW_SFID_SAMPLER, dst, offset, addr,
+ brw_message_desc(devinfo, inst->mlen, 1, inst->header_size) |
+ brw_sampler_desc(devinfo,
+ 0 /* surface */,
+ 0 /* sampler */,
+ GEN5_SAMPLER_MESSAGE_SAMPLE_LD,
+ BRW_SAMPLER_SIMD_MODE_SIMD4X2,
+ 0),
+ false /* EOT */);
}
}
static void
generate_set_simd4x2_header_gen9(struct brw_codegen *p,
- vec4_instruction *inst,
+ vec4_instruction *,
struct brw_reg dst)
{
brw_push_insn_state(p);
static void
generate_mov_indirect(struct brw_codegen *p,
- vec4_instruction *inst,
+ vec4_instruction *,
struct brw_reg dst, struct brw_reg reg,
- struct brw_reg indirect, struct brw_reg length)
+ struct brw_reg indirect)
{
assert(indirect.type == BRW_REGISTER_TYPE_UD);
assert(p->devinfo->gen >= 6);
case SHADER_OPCODE_SHADER_TIME_ADD:
brw_shader_time_add(p, src[0],
prog_data->base.binding_table.shader_time_start);
- brw_mark_surface_used(&prog_data->base,
- prog_data->base.binding_table.shader_time_start);
break;
- case SHADER_OPCODE_UNTYPED_ATOMIC:
+ case VEC4_OPCODE_UNTYPED_ATOMIC:
assert(src[2].file == BRW_IMMEDIATE_VALUE);
brw_untyped_atomic(p, dst, src[0], src[1], src[2].ud, inst->mlen,
- !inst->dst.is_null());
+ !inst->dst.is_null(), inst->header_size);
break;
- case SHADER_OPCODE_UNTYPED_SURFACE_READ:
+ case VEC4_OPCODE_UNTYPED_SURFACE_READ:
+ assert(!inst->header_size);
assert(src[2].file == BRW_IMMEDIATE_VALUE);
brw_untyped_surface_read(p, dst, src[0], src[1], inst->mlen,
src[2].ud);
break;
- case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
+ case VEC4_OPCODE_UNTYPED_SURFACE_WRITE:
assert(src[2].file == BRW_IMMEDIATE_VALUE);
brw_untyped_surface_write(p, src[0], src[1], inst->mlen,
- src[2].ud);
- break;
-
- case SHADER_OPCODE_TYPED_ATOMIC:
- assert(src[2].file == BRW_IMMEDIATE_VALUE);
- brw_typed_atomic(p, dst, src[0], src[1], src[2].ud, inst->mlen,
- !inst->dst.is_null());
- break;
-
- case SHADER_OPCODE_TYPED_SURFACE_READ:
- assert(src[2].file == BRW_IMMEDIATE_VALUE);
- brw_typed_surface_read(p, dst, src[0], src[1], inst->mlen,
- src[2].ud);
- break;
-
- case SHADER_OPCODE_TYPED_SURFACE_WRITE:
- assert(src[2].file == BRW_IMMEDIATE_VALUE);
- brw_typed_surface_write(p, src[0], src[1], inst->mlen,
- src[2].ud);
+ src[2].ud, inst->header_size);
break;
case SHADER_OPCODE_MEMORY_FENCE:
- brw_memory_fence(p, dst);
+ brw_memory_fence(p, dst, BRW_OPCODE_SEND);
break;
case SHADER_OPCODE_FIND_LIVE_CHANNEL: {
break;
case SHADER_OPCODE_MOV_INDIRECT:
- generate_mov_indirect(p, inst, dst, src[0], src[1], src[2]);
+ generate_mov_indirect(p, inst, dst, src[0], src[1]);
break;
case BRW_OPCODE_DIM:
void *mem_ctx,
const nir_shader *nir,
struct brw_vue_prog_data *prog_data,
- const struct cfg_t *cfg,
- unsigned *out_assembly_size)
+ const struct cfg_t *cfg)
{
struct brw_codegen *p = rzalloc(mem_ctx, struct brw_codegen);
brw_init_codegen(compiler->devinfo, p, mem_ctx);
generate_code(p, compiler, log_data, nir, prog_data, cfg);
- return brw_get_program(p, out_assembly_size);
+ return brw_get_program(p, &prog_data->base.program_size);
}