* native instructions.
*/
-#include "main/macros.h"
-#include "brw_context.h"
#include "brw_eu.h"
#include "brw_fs.h"
#include "brw_cfg.h"
+#include "brw_program.h"
-static uint32_t brw_file_from_reg(fs_reg *reg)
+static enum brw_reg_file
+brw_file_from_reg(fs_reg *reg)
{
switch (reg->file) {
- case GRF:
+ case ARF:
+ return BRW_ARCHITECTURE_REGISTER_FILE;
+ case FIXED_GRF:
+ case VGRF:
return BRW_GENERAL_REGISTER_FILE;
case MRF:
return BRW_MESSAGE_REGISTER_FILE;
case IMM:
return BRW_IMMEDIATE_VALUE;
case BAD_FILE:
- case HW_REG:
case ATTR:
case UNIFORM:
unreachable("not reached");
}
- return 0;
+ return BRW_ARCHITECTURE_REGISTER_FILE;
}
static struct brw_reg
switch (reg->file) {
case MRF:
- assert((reg->reg & ~(1 << 7)) < BRW_MAX_MRF(gen));
+ assert((reg->nr & ~BRW_MRF_COMPR4) < BRW_MAX_MRF(gen));
/* Fallthrough */
- case GRF:
+ case VGRF:
if (reg->stride == 0) {
- brw_reg = brw_vec1_reg(brw_file_from_reg(reg), reg->reg, 0);
+ brw_reg = brw_vec1_reg(brw_file_from_reg(reg), reg->nr, 0);
} else if (inst->exec_size < 8) {
- brw_reg = brw_vec8_reg(brw_file_from_reg(reg), reg->reg, 0);
+ brw_reg = brw_vec8_reg(brw_file_from_reg(reg), reg->nr, 0);
brw_reg = stride(brw_reg, inst->exec_size * reg->stride,
inst->exec_size, reg->stride);
} else {
* So, for registers with width > 8, we have to use a width of 8
* and trust the compression state to sort out the exec size.
*/
- brw_reg = brw_vec8_reg(brw_file_from_reg(reg), reg->reg, 0);
+ brw_reg = brw_vec8_reg(brw_file_from_reg(reg), reg->nr, 0);
brw_reg = stride(brw_reg, 8 * reg->stride, 8, reg->stride);
}
brw_reg = retype(brw_reg, reg->type);
brw_reg = byte_offset(brw_reg, reg->subreg_offset);
+ brw_reg.abs = reg->abs;
+ brw_reg.negate = reg->negate;
break;
+ case ARF:
+ case FIXED_GRF:
case IMM:
- assert(reg->stride == ((reg->type == BRW_REGISTER_TYPE_V ||
- reg->type == BRW_REGISTER_TYPE_UV ||
- reg->type == BRW_REGISTER_TYPE_VF) ? 1 : 0));
-
- switch (reg->type) {
- case BRW_REGISTER_TYPE_F:
- brw_reg = brw_imm_f(reg->fixed_hw_reg.dw1.f);
- break;
- case BRW_REGISTER_TYPE_D:
- brw_reg = brw_imm_d(reg->fixed_hw_reg.dw1.d);
- break;
- case BRW_REGISTER_TYPE_UD:
- brw_reg = brw_imm_ud(reg->fixed_hw_reg.dw1.ud);
- break;
- case BRW_REGISTER_TYPE_W:
- brw_reg = brw_imm_w(reg->fixed_hw_reg.dw1.d);
- break;
- case BRW_REGISTER_TYPE_UW:
- brw_reg = brw_imm_uw(reg->fixed_hw_reg.dw1.ud);
- break;
- case BRW_REGISTER_TYPE_VF:
- brw_reg = brw_imm_vf(reg->fixed_hw_reg.dw1.ud);
- break;
- default:
- unreachable("not reached");
- }
- break;
- case HW_REG:
- assert(reg->type == reg->fixed_hw_reg.type);
- brw_reg = reg->fixed_hw_reg;
+ brw_reg = reg->as_brw_reg();
break;
case BAD_FILE:
/* Probably unused. */
case UNIFORM:
unreachable("not reached");
}
- if (reg->abs)
- brw_reg = brw_abs(brw_reg);
- if (reg->negate)
- brw_reg = negate(brw_reg);
return brw_reg;
}
struct brw_stage_prog_data *prog_data,
unsigned promoted_constants,
bool runtime_check_aads_emit,
- const char *stage_abbrev)
+ gl_shader_stage stage)
: compiler(compiler), log_data(log_data),
devinfo(compiler->devinfo), key(key),
prog_data(prog_data),
promoted_constants(promoted_constants),
runtime_check_aads_emit(runtime_check_aads_emit), debug_flag(false),
- stage_abbrev(stage_abbrev), mem_ctx(mem_ctx)
+ stage(stage), mem_ctx(mem_ctx)
{
p = rzalloc(mem_ctx, struct brw_codegen);
brw_init_codegen(devinfo, p, mem_ctx);
}
}
+void
+fs_generator::generate_mov_indirect(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg reg,
+ struct brw_reg indirect_byte_offset)
+{
+ assert(indirect_byte_offset.type == BRW_REGISTER_TYPE_UD);
+ assert(indirect_byte_offset.file == BRW_GENERAL_REGISTER_FILE);
+
+ unsigned imm_byte_offset = reg.nr * REG_SIZE + reg.subnr;
+
+ if (indirect_byte_offset.file == BRW_IMMEDIATE_VALUE) {
+ imm_byte_offset += indirect_byte_offset.ud;
+
+ reg.nr = imm_byte_offset / REG_SIZE;
+ reg.subnr = imm_byte_offset % REG_SIZE;
+ brw_MOV(p, dst, reg);
+ } else {
+ /* Prior to Broadwell, there are only 8 address registers. */
+ assert(inst->exec_size == 8 || devinfo->gen >= 8);
+
+ /* We use VxH indirect addressing, clobbering a0.0 through a0.7. */
+ struct brw_reg addr = vec8(brw_address_reg(0));
+
+ /* The destination stride of an instruction (in bytes) must be greater
+ * than or equal to the size of the rest of the instruction. Since the
+ * address register is of type UW, we can't use a D-type instruction.
+ * In order to get around this, re re-type to UW and use a stride.
+ */
+ indirect_byte_offset =
+ retype(spread(indirect_byte_offset, 2), BRW_REGISTER_TYPE_UW);
+
+ if (devinfo->gen < 8) {
+ /* Prior to broadwell, we have a restriction that the bottom 5 bits
+ * of the base offset and the bottom 5 bits of the indirect must add
+ * to less than 32. In other words, the hardware needs to be able to
+ * add the bottom five bits of the two to get the subnumber and add
+ * the next 7 bits of each to get the actual register number. Since
+ * the indirect may cause us to cross a register boundary, this makes
+ * it almost useless. We could try and do something clever where we
+ * use a actual base offset if base_offset % 32 == 0 but that would
+ * mean we were generating different code depending on the base
+ * offset. Instead, for the sake of consistency, we'll just do the
+ * add ourselves.
+ */
+ brw_ADD(p, addr, indirect_byte_offset, brw_imm_uw(imm_byte_offset));
+ brw_MOV(p, dst, retype(brw_VxH_indirect(0, 0), dst.type));
+ } else {
+ brw_MOV(p, addr, indirect_byte_offset);
+ brw_MOV(p, dst, retype(brw_VxH_indirect(0, imm_byte_offset), dst.type));
+ }
+ }
+}
+
void
fs_generator::generate_urb_read(fs_inst *inst,
struct brw_reg dst,
brw_inst_set_sfid(p->devinfo, send, BRW_SFID_URB);
brw_inst_set_urb_opcode(p->devinfo, send, GEN8_URB_OPCODE_SIMD8_READ);
+ if (inst->opcode == SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT)
+ brw_inst_set_urb_per_slot_offset(p->devinfo, send, true);
+
brw_inst_set_mlen(p->devinfo, send, inst->mlen);
brw_inst_set_rlen(p->devinfo, send, inst->regs_written);
brw_inst_set_header_present(p->devinfo, send, true);
retype(dst, BRW_REGISTER_TYPE_UW),
inst->base_mrf,
src,
- surf_index.dw1.ud,
+ surf_index.ud,
0,
GEN5_SAMPLER_MESSAGE_SAMPLE_RESINFO,
rlen, /* response length */
inst->header_size > 0,
simd_mode,
BRW_SAMPLER_RETURN_FORMAT_SINT32);
+
+ brw_mark_surface_used(prog_data, surf_index.ud);
}
void
fs_generator::generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
+ struct brw_reg surface_index,
struct brw_reg sampler_index)
{
int msg_type = -1;
break;
}
+ /* Stomp the resinfo output type to UINT32. On gens 4-5, the output type
+ * is set as part of the message descriptor. On gen4, the PRM seems to
+ * allow UINT32 and FLOAT32 (i965 PRM, Vol. 4 Section 4.8.1.1), but on
+ * later gens UINT32 is required. Once you hit Sandy Bridge, the bit is
+ * gone from the message descriptor entirely and you just get UINT32 all
+ * the time regasrdless. Since we can really only do non-UINT32 on gen4,
+ * just stomp it to UINT32 all the time.
+ */
+ if (inst->opcode == SHADER_OPCODE_TXS)
+ return_format = BRW_SAMPLER_RETURN_FORMAT_UINT32;
+
switch (inst->exec_size) {
case 8:
simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD8;
/* Set the offset bits in DWord 2. */
brw_MOV(p, get_element_ud(header_reg, 2),
brw_imm_ud(inst->offset));
+ } else if (stage != MESA_SHADER_VERTEX &&
+ stage != MESA_SHADER_FRAGMENT) {
+ /* The vertex and fragment stages have g0.2 set to 0, so
+ * header0.2 is 0 when g0 is copied. Other stages may not, so we
+ * must set it to 0 to avoid setting undesirable bits in the
+ * message.
+ */
+ brw_MOV(p, get_element_ud(header_reg, 2), brw_imm_ud(0));
}
brw_adjust_sampler_state_pointer(p, header_reg, sampler_index);
? prog_data->binding_table.gather_texture_start
: prog_data->binding_table.texture_start;
- if (sampler_index.file == BRW_IMMEDIATE_VALUE) {
- uint32_t sampler = sampler_index.dw1.ud;
+ if (surface_index.file == BRW_IMMEDIATE_VALUE &&
+ sampler_index.file == BRW_IMMEDIATE_VALUE) {
+ uint32_t surface = surface_index.ud;
+ uint32_t sampler = sampler_index.ud;
brw_SAMPLE(p,
retype(dst, BRW_REGISTER_TYPE_UW),
inst->base_mrf,
src,
- sampler + base_binding_table_index,
+ surface + base_binding_table_index,
sampler % 16,
msg_type,
rlen,
simd_mode,
return_format);
- brw_mark_surface_used(prog_data, sampler + base_binding_table_index);
+ brw_mark_surface_used(prog_data, surface + base_binding_table_index);
} else {
/* Non-const sampler index */
struct brw_reg addr = vec1(retype(brw_address_reg(0), BRW_REGISTER_TYPE_UD));
+ struct brw_reg surface_reg = vec1(retype(surface_index, BRW_REGISTER_TYPE_UD));
struct brw_reg sampler_reg = vec1(retype(sampler_index, BRW_REGISTER_TYPE_UD));
brw_push_insn_state(p);
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_set_default_access_mode(p, BRW_ALIGN_1);
- /* addr = ((sampler * 0x101) + base_binding_table_index) & 0xfff */
- brw_MUL(p, addr, sampler_reg, brw_imm_uw(0x101));
+ if (memcmp(&surface_reg, &sampler_reg, sizeof(surface_reg)) == 0) {
+ brw_MUL(p, addr, sampler_reg, brw_imm_uw(0x101));
+ } else {
+ brw_SHL(p, addr, sampler_reg, brw_imm_ud(8));
+ brw_OR(p, addr, addr, surface_reg);
+ }
if (base_binding_table_index)
brw_ADD(p, addr, addr, brw_imm_ud(base_binding_table_index));
brw_AND(p, addr, addr, brw_imm_ud(0xfff));
assert(index.file == BRW_IMMEDIATE_VALUE &&
index.type == BRW_REGISTER_TYPE_UD);
- uint32_t surf_index = index.dw1.ud;
+ uint32_t surf_index = index.ud;
assert(offset.file == BRW_IMMEDIATE_VALUE &&
offset.type == BRW_REGISTER_TYPE_UD);
- uint32_t read_offset = offset.dw1.ud;
+ uint32_t read_offset = offset.ud;
brw_oword_block_read(p, dst, brw_message_reg(inst->base_mrf),
read_offset, surf_index);
if (index.file == BRW_IMMEDIATE_VALUE) {
- uint32_t surf_index = index.dw1.ud;
+ uint32_t surf_index = index.ud;
brw_push_insn_state(p);
brw_set_default_compression_control(p, BRW_COMPRESSION_NONE);
assert(index.file == BRW_IMMEDIATE_VALUE &&
index.type == BRW_REGISTER_TYPE_UD);
- uint32_t surf_index = index.dw1.ud;
+ uint32_t surf_index = index.ud;
uint32_t simd_mode, rlen, msg_type;
if (dispatch_width == 16) {
if (index.file == BRW_IMMEDIATE_VALUE) {
- uint32_t surf_index = index.dw1.ud;
+ uint32_t surf_index = index.ud;
brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
brw_set_dest(p, send, retype(dst, BRW_REGISTER_TYPE_UW));
case SHADER_OPCODE_TG4:
case SHADER_OPCODE_TG4_OFFSET:
case SHADER_OPCODE_SAMPLEINFO:
- generate_tex(inst, dst, src[0], src[1]);
+ generate_tex(inst, dst, src[0], src[1], src[2]);
break;
case FS_OPCODE_DDX_COARSE:
case FS_OPCODE_DDX_FINE:
case FS_OPCODE_DDY_COARSE:
case FS_OPCODE_DDY_FINE:
assert(src[1].file == BRW_IMMEDIATE_VALUE);
- generate_ddy(inst->opcode, dst, src[0], src[1].dw1.ud);
+ generate_ddy(inst->opcode, dst, src[0], src[1].ud);
break;
case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
fill_count++;
break;
+ case SHADER_OPCODE_MOV_INDIRECT:
+ generate_mov_indirect(inst, dst, src[0], src[1]);
+ break;
+
case SHADER_OPCODE_URB_READ_SIMD8:
+ case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
generate_urb_read(inst, dst, src[0]);
break;
case SHADER_OPCODE_UNTYPED_ATOMIC:
assert(src[2].file == BRW_IMMEDIATE_VALUE);
- brw_untyped_atomic(p, dst, src[0], src[1], src[2].dw1.ud,
+ brw_untyped_atomic(p, dst, src[0], src[1], src[2].ud,
inst->mlen, !inst->dst.is_null());
break;
case SHADER_OPCODE_UNTYPED_SURFACE_READ:
assert(src[2].file == BRW_IMMEDIATE_VALUE);
brw_untyped_surface_read(p, dst, src[0], src[1],
- inst->mlen, src[2].dw1.ud);
+ inst->mlen, src[2].ud);
break;
case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
assert(src[2].file == BRW_IMMEDIATE_VALUE);
brw_untyped_surface_write(p, src[0], src[1],
- inst->mlen, src[2].dw1.ud);
+ 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].dw1.ud, inst->mlen, !inst->dst.is_null());
+ 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].dw1.ud);
+ 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].dw1.ud);
+ brw_typed_surface_write(p, src[0], src[1], inst->mlen, src[2].ud);
break;
case SHADER_OPCODE_MEMORY_FENCE:
brw_set_uip_jip(p);
annotation_finalize(&annotation, p->next_insn_offset);
+#ifndef NDEBUG
+ bool validated = brw_validate_instructions(p, start_offset, &annotation);
+#else
+ if (unlikely(debug_flag))
+ brw_validate_instructions(p, start_offset, &annotation);
+#endif
+
int before_size = p->next_insn_offset - start_offset;
brw_compact_instructions(p, start_offset, annotation.ann_count,
annotation.ann);
p->devinfo);
ralloc_free(annotation.mem_ctx);
}
+ assert(validated);
compiler->shader_debug_log(log_data,
"%s SIMD%d shader: %d inst, %d loops, %u cycles, "
"%d:%d spills:fills, Promoted %u constants, "
- "compacted %d to %d bytes.\n",
- stage_abbrev, dispatch_width, before_size / 16,
+ "compacted %d to %d bytes.",
+ _mesa_shader_stage_to_abbrev(stage),
+ dispatch_width, before_size / 16,
loop_count, cfg->cycle_count, spill_count,
fill_count, promoted_constants, before_size,
after_size);
projects / mesa.git / blobdiff