* IN THE SOFTWARE.
*/
-#include "glsl/glsl_parser_extras.h"
#include "brw_vec4.h"
#include "brw_cfg.h"
+#include "brw_eu.h"
+#include "brw_program.h"
using namespace brw;
struct brw_reg surface_index,
struct brw_reg sampler_index)
{
- const struct brw_device_info *devinfo = p->devinfo;
+ const struct gen_device_info *devinfo = p->devinfo;
int msg_type = -1;
if (devinfo->gen >= 5) {
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
brw_set_default_access_mode(p, BRW_ALIGN_1);
- if (memcmp(&surface_reg, &sampler_reg, sizeof(surface_reg)) == 0) {
+ if (brw_regs_equal(&surface_reg, &sampler_reg)) {
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 (sampler_reg.file == BRW_IMMEDIATE_VALUE) {
+ brw_OR(p, addr, surface_reg, brw_imm_ud(sampler_reg.ud << 8));
+ } 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));
/* We pass the temporary passed in src0 as the writeback register */
brw_urb_WRITE(p,
- inst->src[0], /* dest */
+ inst->src[0].as_brw_reg(), /* dest */
inst->base_mrf, /* starting mrf reg nr */
src,
BRW_URB_WRITE_ALLOCATE_COMPLETE,
brw_push_insn_state(p);
brw_set_default_access_mode(p, BRW_ALIGN_1);
brw_set_default_mask_control(p, BRW_MASK_DISABLE);
- brw_MOV(p, get_element_ud(inst->dst, 0),
- get_element_ud(inst->src[0], 0));
+ brw_MOV(p, get_element_ud(inst->dst.as_brw_reg(), 0),
+ get_element_ud(inst->src[0].as_brw_reg(), 0));
brw_pop_insn_state(p);
}
bool final_write = inst->sol_final_write;
brw_push_insn_state(p);
+ brw_set_default_exec_size(p, BRW_EXECUTE_4);
/* Copy Vertex data into M0.x */
brw_MOV(p, stride(dst, 4, 4, 1),
stride(retype(src0, BRW_REGISTER_TYPE_UD), 4, 4, 1));
+ brw_pop_insn_state(p);
+ brw_push_insn_state(p);
/* Send SVB Write */
brw_svb_write(p,
final_write ? src1 : brw_null_reg(), /* dest == src1 */
brw_MOV(p, get_element_ud(header, 0), get_element_ud(dst, 0));
/* src1 is not an immediate when we use transform feedback */
- if (src1.file != BRW_IMMEDIATE_VALUE)
+ if (src1.file != BRW_IMMEDIATE_VALUE) {
+ brw_set_default_exec_size(p, BRW_EXECUTE_4);
brw_MOV(p, brw_vec4_grf(src1.nr, 0), brw_vec4_grf(dst.nr, 1));
+ }
brw_pop_insn_state(p);
}
brw_pop_insn_state(p);
}
+static void
+generate_tcs_get_instance_id(struct brw_codegen *p, struct brw_reg dst)
+{
+ const struct gen_device_info *devinfo = p->devinfo;
+ const bool ivb = devinfo->is_ivybridge || devinfo->is_baytrail;
+
+ /* "Instance Count" comes as part of the payload in r0.2 bits 23:17.
+ *
+ * Since we operate in SIMD4x2 mode, we need run half as many threads
+ * as necessary. So we assign (2i + 1, 2i) as the thread counts. We
+ * shift right by one less to accomplish the multiplication by two.
+ */
+ dst = retype(dst, BRW_REGISTER_TYPE_UD);
+ struct brw_reg r0(retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
+
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+
+ const int mask = ivb ? INTEL_MASK(22, 16) : INTEL_MASK(23, 17);
+ const int shift = ivb ? 16 : 17;
+
+ brw_AND(p, get_element_ud(dst, 0), get_element_ud(r0, 2), brw_imm_ud(mask));
+ brw_SHR(p, get_element_ud(dst, 0), get_element_ud(dst, 0),
+ brw_imm_ud(shift - 1));
+ brw_ADD(p, get_element_ud(dst, 4), get_element_ud(dst, 0), brw_imm_ud(1));
+
+ brw_pop_insn_state(p);
+}
+
+static void
+generate_tcs_urb_write(struct brw_codegen *p,
+ vec4_instruction *inst,
+ struct brw_reg urb_header)
+{
+ const struct gen_device_info *devinfo = p->devinfo;
+
+ 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_message_descriptor(p, send, BRW_SFID_URB,
+ inst->mlen /* mlen */, 0 /* rlen */,
+ true /* header */, false /* eot */);
+ 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_inst_set_eot(devinfo, send, 1);
+ } else {
+ brw_inst_set_urb_per_slot_offset(devinfo, send, 1);
+ brw_inst_set_urb_swizzle_control(devinfo, send, BRW_URB_SWIZZLE_INTERLEAVE);
+ }
+
+ /* what happens to swizzles? */
+}
+
+
+static void
+generate_tcs_input_urb_offsets(struct brw_codegen *p,
+ struct brw_reg dst,
+ struct brw_reg vertex,
+ struct brw_reg offset)
+{
+ /* Generates an URB read/write message header for HS/DS operation.
+ * Inputs are a vertex index, and a byte offset from the beginning of
+ * the vertex. */
+
+ /* If `vertex` is not an immediate, we clobber a0.0 */
+
+ assert(vertex.file == BRW_IMMEDIATE_VALUE || vertex.file == BRW_GENERAL_REGISTER_FILE);
+ assert(vertex.type == BRW_REGISTER_TYPE_UD || vertex.type == BRW_REGISTER_TYPE_D);
+
+ assert(dst.file == BRW_GENERAL_REGISTER_FILE);
+
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+ brw_MOV(p, dst, brw_imm_ud(0));
+
+ /* m0.5 bits 8-15 are channel enables */
+ brw_MOV(p, get_element_ud(dst, 5), brw_imm_ud(0xff00));
+
+ /* m0.0-0.1: URB handles */
+ if (vertex.file == BRW_IMMEDIATE_VALUE) {
+ uint32_t vertex_index = vertex.ud;
+ struct brw_reg index_reg = brw_vec1_grf(
+ 1 + (vertex_index >> 3), vertex_index & 7);
+
+ brw_MOV(p, vec2(get_element_ud(dst, 0)),
+ retype(index_reg, BRW_REGISTER_TYPE_UD));
+ } else {
+ /* Use indirect addressing. ICP Handles are DWords (single channels
+ * of a register) and start at g1.0.
+ *
+ * In order to start our region at g1.0, we add 8 to the vertex index,
+ * effectively skipping over the 8 channels in g0.0. This gives us a
+ * DWord offset to the ICP Handle.
+ *
+ * Indirect addressing works in terms of bytes, so we then multiply
+ * the DWord offset by 4 (by shifting left by 2).
+ */
+ struct brw_reg addr = brw_address_reg(0);
+
+ /* bottom half: m0.0 = g[1.0 + vertex.0]UD */
+ brw_ADD(p, addr, get_element_ud(vertex, 0), brw_imm_uw(0x8));
+ brw_SHL(p, addr, addr, brw_imm_ud(2));
+ brw_MOV(p, get_element_ud(dst, 0), deref_1ud(brw_indirect(0, 0), 0));
+
+ /* top half: m0.1 = g[1.0 + vertex.4]UD */
+ brw_ADD(p, addr, get_element_ud(vertex, 4), brw_imm_uw(0x8));
+ brw_SHL(p, addr, addr, brw_imm_ud(2));
+ brw_MOV(p, get_element_ud(dst, 1), deref_1ud(brw_indirect(0, 0), 0));
+ }
+
+ /* m0.3-0.4: 128bit-granular offsets into the URB from the handles */
+ if (offset.file != ARF)
+ brw_MOV(p, vec2(get_element_ud(dst, 3)), stride(offset, 4, 1, 0));
+
+ brw_pop_insn_state(p);
+}
+
+
+static void
+generate_tcs_output_urb_offsets(struct brw_codegen *p,
+ struct brw_reg dst,
+ struct brw_reg write_mask,
+ struct brw_reg offset)
+{
+ /* Generates an URB read/write message header for HS/DS operation, for the patch URB entry. */
+ assert(dst.file == BRW_GENERAL_REGISTER_FILE || dst.file == BRW_MESSAGE_REGISTER_FILE);
+
+ assert(write_mask.file == BRW_IMMEDIATE_VALUE);
+ assert(write_mask.type == BRW_REGISTER_TYPE_UD);
+
+ brw_push_insn_state(p);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+ brw_MOV(p, dst, brw_imm_ud(0));
+
+ unsigned mask = write_mask.ud;
+
+ /* m0.5 bits 15:12 and 11:8 are channel enables */
+ brw_MOV(p, get_element_ud(dst, 5), brw_imm_ud((mask << 8) | (mask << 12)));
+
+ /* HS patch URB handle is delivered in r0.0 */
+ struct brw_reg urb_handle = brw_vec1_grf(0, 0);
+
+ /* m0.0-0.1: URB handles */
+ brw_MOV(p, vec2(get_element_ud(dst, 0)),
+ retype(urb_handle, BRW_REGISTER_TYPE_UD));
+
+ /* m0.3-0.4: 128bit-granular offsets into the URB from the handles */
+ if (offset.file != ARF)
+ brw_MOV(p, vec2(get_element_ud(dst, 3)), stride(offset, 4, 1, 0));
+
+ brw_pop_insn_state(p);
+}
+
+static void
+generate_tes_create_input_read_header(struct brw_codegen *p,
+ struct brw_reg dst)
+{
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+
+ /* Initialize the register to 0 */
+ brw_MOV(p, dst, brw_imm_ud(0));
+
+ /* Enable all the channels in m0.5 bits 15:8 */
+ brw_MOV(p, get_element_ud(dst, 5), brw_imm_ud(0xff00));
+
+ /* Copy g1.3 (the patch URB handle) to m0.0 and m0.1. For safety,
+ * mask out irrelevant "Reserved" bits, as they're not marked MBZ.
+ */
+ brw_AND(p, vec2(get_element_ud(dst, 0)),
+ retype(brw_vec1_grf(1, 3), BRW_REGISTER_TYPE_UD),
+ brw_imm_ud(0x1fff));
+ brw_pop_insn_state(p);
+}
+
+static void
+generate_tes_add_indirect_urb_offset(struct brw_codegen *p,
+ struct brw_reg dst,
+ struct brw_reg header,
+ struct brw_reg offset)
+{
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+
+ brw_MOV(p, dst, header);
+ /* 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_pop_insn_state(p);
+}
+
+static void
+generate_vec4_urb_read(struct brw_codegen *p,
+ vec4_instruction *inst,
+ struct brw_reg dst,
+ struct brw_reg header)
+{
+ const struct gen_device_info *devinfo = p->devinfo;
+
+ assert(header.file == BRW_GENERAL_REGISTER_FILE);
+ assert(header.type == BRW_REGISTER_TYPE_UD);
+
+ brw_inst *send = brw_next_insn(p, BRW_OPCODE_SEND);
+ 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_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_set_urb_global_offset(devinfo, send, inst->offset);
+}
+
+static void
+generate_tcs_release_input(struct brw_codegen *p,
+ struct brw_reg header,
+ struct brw_reg vertex,
+ struct brw_reg is_unpaired)
+{
+ const struct gen_device_info *devinfo = p->devinfo;
+
+ assert(vertex.file == BRW_IMMEDIATE_VALUE);
+ assert(vertex.type == BRW_REGISTER_TYPE_UD);
+
+ /* m0.0-0.1: URB handles */
+ struct brw_reg urb_handles =
+ retype(brw_vec2_grf(1 + (vertex.ud >> 3), vertex.ud & 7),
+ BRW_REGISTER_TYPE_UD);
+
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+ brw_MOV(p, header, brw_imm_ud(0));
+ brw_MOV(p, vec2(get_element_ud(header, 0)), urb_handles);
+ brw_pop_insn_state(p);
+
+ 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_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 ?
+ BRW_URB_SWIZZLE_NONE :
+ BRW_URB_SWIZZLE_INTERLEAVE);
+}
+
+static void
+generate_tcs_thread_end(struct brw_codegen *p, vec4_instruction *inst)
+{
+ struct brw_reg header = brw_message_reg(inst->base_mrf);
+
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+ brw_MOV(p, header, brw_imm_ud(0));
+ brw_MOV(p, get_element_ud(header, 5), brw_imm_ud(WRITEMASK_X << 8));
+ brw_MOV(p, get_element_ud(header, 0),
+ retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD));
+ brw_MOV(p, brw_message_reg(inst->base_mrf + 1), brw_imm_ud(0u));
+ brw_pop_insn_state(p);
+
+ brw_urb_WRITE(p,
+ brw_null_reg(), /* dest */
+ inst->base_mrf, /* starting mrf reg nr */
+ header,
+ BRW_URB_WRITE_EOT | BRW_URB_WRITE_OWORD |
+ BRW_URB_WRITE_USE_CHANNEL_MASKS,
+ inst->mlen,
+ 0, /* response len */
+ 0, /* urb destination offset */
+ 0);
+}
+
+static void
+generate_tes_get_primitive_id(struct brw_codegen *p, struct brw_reg dst)
+{
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_MOV(p, dst, retype(brw_vec1_grf(1, 7), BRW_REGISTER_TYPE_D));
+ brw_pop_insn_state(p);
+}
+
+static void
+generate_tcs_get_primitive_id(struct brw_codegen *p, struct brw_reg dst)
+{
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_MOV(p, dst, retype(brw_vec1_grf(0, 1), BRW_REGISTER_TYPE_UD));
+ brw_pop_insn_state(p);
+}
+
+static void
+generate_tcs_create_barrier_header(struct brw_codegen *p,
+ struct brw_vue_prog_data *prog_data,
+ struct brw_reg dst)
+{
+ const struct gen_device_info *devinfo = p->devinfo;
+ const bool ivb = devinfo->is_ivybridge || devinfo->is_baytrail;
+ struct brw_reg m0_2 = get_element_ud(dst, 2);
+ unsigned instances = ((struct brw_tcs_prog_data *) prog_data)->instances;
+
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+
+ /* Zero the message header */
+ brw_MOV(p, retype(dst, BRW_REGISTER_TYPE_UD), brw_imm_ud(0u));
+
+ /* Copy "Barrier ID" from r0.2, bits 16:13 (Gen7.5+) or 15:12 (Gen7) */
+ brw_AND(p, m0_2,
+ retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_UD),
+ brw_imm_ud(ivb ? INTEL_MASK(15, 12) : INTEL_MASK(16, 13)));
+
+ /* Shift it up to bits 27:24. */
+ brw_SHL(p, m0_2, get_element_ud(dst, 2), brw_imm_ud(ivb ? 12 : 11));
+
+ /* Set the Barrier Count and the enable bit */
+ brw_OR(p, m0_2, m0_2, brw_imm_ud(instances << 9 | (1 << 15)));
+
+ brw_pop_insn_state(p);
+}
+
static void
generate_oword_dual_block_offsets(struct brw_codegen *p,
struct brw_reg m1,
struct brw_reg dst,
struct brw_reg index)
{
- const struct brw_device_info *devinfo = p->devinfo;
+ const struct gen_device_info *devinfo = p->devinfo;
struct brw_reg header = brw_vec8_grf(0, 0);
gen6_resolve_implied_move(p, &header, inst->base_mrf);
else
msg_type = BRW_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ;
+ 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;
+
/* Each of the 8 channel enables is considered for whether each
* dword is written.
*/
if (devinfo->gen < 6)
brw_inst_set_cond_modifier(devinfo, send, inst->base_mrf);
brw_set_dp_read_message(p, send,
- 255, /* binding table index: stateless access */
+ brw_scratch_surface_idx(p),
BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
- msg_type,
- BRW_DATAPORT_READ_TARGET_RENDER_CACHE,
+ msg_type, target_cache,
2, /* mlen */
true, /* header_present */
1 /* rlen */);
struct brw_reg src,
struct brw_reg index)
{
- const struct brw_device_info *devinfo = p->devinfo;
+ const struct gen_device_info *devinfo = p->devinfo;
+ 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);
struct brw_reg header = brw_vec8_grf(0, 0);
bool write_commit;
if (devinfo->gen < 6)
brw_inst_set_cond_modifier(p->devinfo, send, inst->base_mrf);
brw_set_dp_write_message(p, send,
- 255, /* binding table index: stateless access */
+ 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 */
struct brw_reg index,
struct brw_reg offset)
{
- const struct brw_device_info *devinfo = p->devinfo;
+ 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);
assert(index.file == BRW_IMMEDIATE_VALUE &&
index.type == BRW_REGISTER_TYPE_UD);
uint32_t surf_index = index.ud;
gen6_resolve_implied_move(p, &header, inst->base_mrf);
- brw_MOV(p, retype(brw_message_reg(inst->base_mrf + 1), BRW_REGISTER_TYPE_D),
- offset);
+ if (devinfo->gen >= 6) {
+ if (offset.file == BRW_IMMEDIATE_VALUE) {
+ brw_MOV(p, retype(brw_message_reg(inst->base_mrf + 1),
+ BRW_REGISTER_TYPE_D),
+ brw_imm_d(offset.ud >> 4));
+ } else {
+ brw_SHR(p, retype(brw_message_reg(inst->base_mrf + 1),
+ BRW_REGISTER_TYPE_D),
+ offset, brw_imm_d(4));
+ }
+ } else {
+ brw_MOV(p, retype(brw_message_reg(inst->base_mrf + 1),
+ BRW_REGISTER_TYPE_D),
+ offset);
+ }
uint32_t msg_type;
surf_index,
BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
msg_type,
- BRW_DATAPORT_READ_TARGET_DATA_CACHE,
+ target_cache,
2, /* mlen */
true, /* header_present */
1 /* rlen */);
brw_pop_insn_state(p);
}
+static void
+generate_mov_indirect(struct brw_codegen *p,
+ vec4_instruction *inst,
+ struct brw_reg dst, struct brw_reg reg,
+ struct brw_reg indirect, struct brw_reg length)
+{
+ assert(indirect.type == BRW_REGISTER_TYPE_UD);
+ assert(p->devinfo->gen >= 6);
+
+ unsigned imm_byte_offset = reg.nr * REG_SIZE + reg.subnr * (REG_SIZE / 2);
+
+ /* This instruction acts in align1 mode */
+ assert(dst.writemask == WRITEMASK_XYZW);
+
+ if (indirect.file == BRW_IMMEDIATE_VALUE) {
+ imm_byte_offset += indirect.ud;
+
+ reg.nr = imm_byte_offset / REG_SIZE;
+ reg.subnr = (imm_byte_offset / (REG_SIZE / 2)) % 2;
+ unsigned shift = (imm_byte_offset / 4) % 4;
+ reg.swizzle += BRW_SWIZZLE4(shift, shift, shift, shift);
+
+ brw_MOV(p, dst, reg);
+ } else {
+ brw_push_insn_state(p);
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+ brw_set_default_mask_control(p, BRW_MASK_DISABLE);
+
+ struct brw_reg addr = vec8(brw_address_reg(0));
+
+ /* We need to move the indirect value into the address register. In
+ * order to make things make some sense, we want to respect at least the
+ * X component of the swizzle. In order to do that, we need to convert
+ * the subnr (probably 0) to an align1 subnr and add in the swizzle.
+ */
+ assert(brw_is_single_value_swizzle(indirect.swizzle));
+ indirect.subnr = (indirect.subnr * 4 + BRW_GET_SWZ(indirect.swizzle, 0));
+
+ /* We then use a region of <8,4,0>:uw to pick off the first 2 bytes of
+ * the indirect and splat it out to all four channels of the given half
+ * of a0.
+ */
+ indirect.subnr *= 2;
+ indirect = stride(retype(indirect, BRW_REGISTER_TYPE_UW), 8, 4, 0);
+ brw_ADD(p, addr, indirect, brw_imm_uw(imm_byte_offset));
+
+ /* Now we need to incorporate the swizzle from the source register */
+ if (reg.swizzle != BRW_SWIZZLE_XXXX) {
+ uint32_t uv_swiz = BRW_GET_SWZ(reg.swizzle, 0) << 2 |
+ BRW_GET_SWZ(reg.swizzle, 1) << 6 |
+ BRW_GET_SWZ(reg.swizzle, 2) << 10 |
+ BRW_GET_SWZ(reg.swizzle, 3) << 14;
+ uv_swiz |= uv_swiz << 16;
+
+ brw_ADD(p, addr, addr, brw_imm_uv(uv_swiz));
+ }
+
+ brw_MOV(p, dst, retype(brw_VxH_indirect(0, 0), reg.type));
+
+ brw_pop_insn_state(p);
+ }
+}
+
static void
generate_code(struct brw_codegen *p,
const struct brw_compiler *compiler,
struct brw_vue_prog_data *prog_data,
const struct cfg_t *cfg)
{
- const struct brw_device_info *devinfo = p->devinfo;
+ const struct gen_device_info *devinfo = p->devinfo;
const char *stage_abbrev = _mesa_shader_stage_to_abbrev(nir->stage);
bool debug_flag = INTEL_DEBUG &
intel_debug_flag_for_shader_stage(nir->stage);
struct annotation_info annotation;
memset(&annotation, 0, sizeof(annotation));
+ int spill_count = 0, fill_count = 0;
int loop_count = 0;
foreach_block_and_inst (block, vec4_instruction, inst, cfg) {
annotate(p->devinfo, &annotation, cfg, inst, p->next_insn_offset);
for (unsigned int i = 0; i < 3; i++) {
- src[i] = inst->src[i];
+ src[i] = inst->src[i].as_brw_reg();
}
- dst = inst->dst;
+ dst = inst->dst.as_brw_reg();
brw_set_default_predicate_control(p, inst->predicate);
brw_set_default_predicate_inverse(p, inst->predicate_inverse);
brw_set_default_saturate(p, inst->saturate);
brw_set_default_mask_control(p, inst->force_writemask_all);
brw_set_default_acc_write_control(p, inst->writes_accumulator);
+ brw_set_default_exec_size(p, cvt(inst->exec_size) - 1);
assert(inst->base_mrf + inst->mlen <= BRW_MAX_MRF(devinfo->gen));
assert(inst->mlen <= BRW_MAX_MSG_LENGTH);
unsigned pre_emit_nr_insn = p->nr_insn;
- if (dst.width == BRW_WIDTH_4) {
- /* This happens in attribute fixups for "dual instanced" geometry
- * shaders, since they use attributes that are vec4's. Since the exec
- * width is only 4, it's essential that the caller set
- * force_writemask_all in order to make sure the instruction is executed
- * regardless of which channels are enabled.
- */
- assert(inst->force_writemask_all);
-
- /* Fix up any <8;8,1> or <0;4,1> source registers to <4;4,1> to satisfy
- * the following register region restrictions (from Graphics BSpec:
- * 3D-Media-GPGPU Engine > EU Overview > Registers and Register Regions
- * > Register Region Restrictions)
- *
- * 1. ExecSize must be greater than or equal to Width.
- *
- * 2. If ExecSize = Width and HorzStride != 0, VertStride must be set
- * to Width * HorzStride."
- */
- for (int i = 0; i < 3; i++) {
- if (src[i].file == BRW_GENERAL_REGISTER_FILE)
- src[i] = stride(src[i], 4, 4, 1);
- }
- }
-
switch (inst->opcode) {
case VEC4_OPCODE_UNPACK_UNIFORM:
case BRW_OPCODE_MOV:
/* FBL only supports UD type for dst. */
brw_FBL(p, retype(dst, BRW_REGISTER_TYPE_UD), src[0]);
break;
+ case BRW_OPCODE_LZD:
+ brw_LZD(p, dst, src[0]);
+ break;
case BRW_OPCODE_CBIT:
assert(devinfo->gen >= 7);
/* CBIT only supports UD type for dst. */
case SHADER_OPCODE_TG4:
case SHADER_OPCODE_TG4_OFFSET:
case SHADER_OPCODE_SAMPLEINFO:
- generate_tex(p, prog_data, inst, dst, src[0], src[1], src[1]);
+ generate_tex(p, prog_data, inst, dst, src[0], src[1], src[2]);
break;
case VS_OPCODE_URB_WRITE:
case SHADER_OPCODE_GEN4_SCRATCH_READ:
generate_scratch_read(p, inst, dst, src[0]);
+ fill_count++;
break;
case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
generate_scratch_write(p, inst, dst, src[0], src[1]);
+ spill_count++;
break;
case VS_OPCODE_PULL_CONSTANT_LOAD:
brw_memory_fence(p, dst);
break;
- case SHADER_OPCODE_FIND_LIVE_CHANNEL:
- brw_find_live_channel(p, dst);
+ case SHADER_OPCODE_FIND_LIVE_CHANNEL: {
+ const struct brw_reg mask =
+ brw_stage_has_packed_dispatch(devinfo, nir->stage,
+ &prog_data->base) ? brw_imm_ud(~0u) :
+ brw_dmask_reg();
+ brw_find_live_channel(p, dst, mask);
break;
+ }
case SHADER_OPCODE_BROADCAST:
+ assert(inst->force_writemask_all);
brw_broadcast(p, dst, src[0], src[1]);
break;
break;
}
+ case VEC4_OPCODE_FROM_DOUBLE: {
+ assert(type_sz(src[0].type) == 8);
+ assert(type_sz(dst.type) == 4);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+
+ dst.hstride = BRW_HORIZONTAL_STRIDE_2;
+ dst.width = BRW_WIDTH_4;
+ src[0].vstride = BRW_VERTICAL_STRIDE_4;
+ src[0].width = BRW_WIDTH_4;
+ brw_MOV(p, dst, src[0]);
+
+ struct brw_reg dst_as_src = dst;
+ dst.hstride = BRW_HORIZONTAL_STRIDE_1;
+ dst.width = BRW_WIDTH_8;
+ brw_MOV(p, dst, dst_as_src);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_16);
+ break;
+ }
+
+ case VEC4_OPCODE_TO_DOUBLE: {
+ assert(type_sz(src[0].type) == 4);
+ assert(type_sz(dst.type) == 8);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+
+ struct brw_reg tmp = retype(dst, src[0].type);
+ tmp.hstride = BRW_HORIZONTAL_STRIDE_2;
+ tmp.width = BRW_WIDTH_4;
+ src[0].vstride = BRW_VERTICAL_STRIDE_4;
+ src[0].hstride = BRW_HORIZONTAL_STRIDE_1;
+ src[0].width = BRW_WIDTH_4;
+ brw_MOV(p, tmp, src[0]);
+
+ tmp.vstride = BRW_VERTICAL_STRIDE_8;
+ tmp.hstride = BRW_HORIZONTAL_STRIDE_2;
+ tmp.width = BRW_WIDTH_4;
+ brw_MOV(p, dst, tmp);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_16);
+ break;
+ }
+
+ case VEC4_OPCODE_PICK_LOW_32BIT:
+ case VEC4_OPCODE_PICK_HIGH_32BIT: {
+ /* Stores the low/high 32-bit of each 64-bit element in src[0] into
+ * dst using ALIGN1 mode and a <8,4,2>:UD region on the source.
+ */
+ assert(type_sz(src[0].type) == 8);
+ assert(type_sz(dst.type) == 4);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+
+ dst = retype(dst, BRW_REGISTER_TYPE_UD);
+ dst.hstride = BRW_HORIZONTAL_STRIDE_1;
+
+ src[0] = retype(src[0], BRW_REGISTER_TYPE_UD);
+ if (inst->opcode == VEC4_OPCODE_PICK_HIGH_32BIT)
+ src[0] = suboffset(src[0], 1);
+ src[0].vstride = BRW_VERTICAL_STRIDE_8;
+ src[0].width = BRW_WIDTH_4;
+ src[0].hstride = BRW_HORIZONTAL_STRIDE_2;
+ brw_MOV(p, dst, src[0]);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_16);
+ break;
+ }
+
+ case VEC4_OPCODE_SET_LOW_32BIT:
+ case VEC4_OPCODE_SET_HIGH_32BIT: {
+ /* Reads consecutive 32-bit elements from src[0] and writes
+ * them to the low/high 32-bit of each 64-bit element in dst.
+ */
+ assert(type_sz(src[0].type) == 4);
+ assert(type_sz(dst.type) == 8);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_1);
+
+ dst = retype(dst, BRW_REGISTER_TYPE_UD);
+ if (inst->opcode == VEC4_OPCODE_SET_HIGH_32BIT)
+ dst = suboffset(dst, 1);
+ dst.hstride = BRW_HORIZONTAL_STRIDE_2;
+
+ src[0] = retype(src[0], BRW_REGISTER_TYPE_UD);
+ src[0].vstride = BRW_VERTICAL_STRIDE_4;
+ src[0].width = BRW_WIDTH_4;
+ src[0].hstride = BRW_HORIZONTAL_STRIDE_1;
+ brw_MOV(p, dst, src[0]);
+
+ brw_set_default_access_mode(p, BRW_ALIGN_16);
+ break;
+ }
+
case VEC4_OPCODE_PACK_BYTES: {
/* Is effectively:
*
break;
}
+ case TCS_OPCODE_URB_WRITE:
+ generate_tcs_urb_write(p, inst, src[0]);
+ break;
+
+ case VEC4_OPCODE_URB_READ:
+ generate_vec4_urb_read(p, inst, dst, src[0]);
+ break;
+
+ case TCS_OPCODE_SET_INPUT_URB_OFFSETS:
+ generate_tcs_input_urb_offsets(p, dst, src[0], src[1]);
+ break;
+
+ case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS:
+ generate_tcs_output_urb_offsets(p, dst, src[0], src[1]);
+ break;
+
+ case TCS_OPCODE_GET_INSTANCE_ID:
+ generate_tcs_get_instance_id(p, dst);
+ break;
+
+ case TCS_OPCODE_GET_PRIMITIVE_ID:
+ generate_tcs_get_primitive_id(p, dst);
+ break;
+
+ case TCS_OPCODE_CREATE_BARRIER_HEADER:
+ generate_tcs_create_barrier_header(p, prog_data, dst);
+ break;
+
+ case TES_OPCODE_CREATE_INPUT_READ_HEADER:
+ generate_tes_create_input_read_header(p, dst);
+ break;
+
+ case TES_OPCODE_ADD_INDIRECT_URB_OFFSET:
+ generate_tes_add_indirect_urb_offset(p, dst, src[0], src[1]);
+ break;
+
+ case TES_OPCODE_GET_PRIMITIVE_ID:
+ generate_tes_get_primitive_id(p, dst);
+ break;
+
+ case TCS_OPCODE_SRC0_010_IS_ZERO:
+ /* If src_reg had stride like fs_reg, we wouldn't need this. */
+ brw_MOV(p, brw_null_reg(), stride(src[0], 0, 1, 0));
+ break;
+
+ case TCS_OPCODE_RELEASE_INPUT:
+ generate_tcs_release_input(p, dst, src[0], src[1]);
+ break;
+
+ case TCS_OPCODE_THREAD_END:
+ generate_tcs_thread_end(p, inst);
+ break;
+
+ case SHADER_OPCODE_BARRIER:
+ brw_barrier(p, src[0]);
+ brw_WAIT(p);
+ break;
+
+ case SHADER_OPCODE_MOV_INDIRECT:
+ generate_mov_indirect(p, inst, dst, src[0], src[1], src[2]);
+ break;
+
+ case BRW_OPCODE_DIM:
+ assert(devinfo->is_haswell);
+ assert(src[0].type == BRW_REGISTER_TYPE_DF);
+ assert(dst.type == BRW_REGISTER_TYPE_DF);
+ brw_DIM(p, dst, retype(src[0], BRW_REGISTER_TYPE_F));
+ break;
+
default:
unreachable("Unsupported opcode");
}
}
}
- brw_set_uip_jip(p);
+ brw_set_uip_jip(p, 0);
annotation_finalize(&annotation, p->next_insn_offset);
#ifndef NDEBUG
if (unlikely(debug_flag)) {
fprintf(stderr, "Native code for %s %s shader %s:\n",
- nir->info.label ? nir->info.label : "unnamed",
- _mesa_shader_stage_to_string(nir->stage), nir->info.name);
+ nir->info->label ? nir->info->label : "unnamed",
+ _mesa_shader_stage_to_string(nir->stage), nir->info->name);
- fprintf(stderr, "%s vec4 shader: %d instructions. %d loops. %u cycles."
- "Compacted %d to %d bytes (%.0f%%)\n",
- stage_abbrev,
- before_size / 16, loop_count, cfg->cycle_count, before_size, after_size,
+ fprintf(stderr, "%s vec4 shader: %d instructions. %d loops. %u cycles. %d:%d "
+ "spills:fills. Compacted %d to %d bytes (%.0f%%)\n",
+ stage_abbrev, before_size / 16, loop_count, cfg->cycle_count,
+ spill_count, fill_count, before_size, after_size,
100.0f * (before_size - after_size) / before_size);
dump_assembly(p->store, annotation.ann_count, annotation.ann,
compiler->shader_debug_log(log_data,
"%s vec4 shader: %d inst, %d loops, %u cycles, "
- "compacted %d to %d bytes.\n",
+ "%d:%d spills:fills, compacted %d to %d bytes.",
stage_abbrev, before_size / 16,
- loop_count, cfg->cycle_count,
- before_size, after_size);
+ loop_count, cfg->cycle_count, spill_count,
+ fill_count, before_size, after_size);
+
}
extern "C" const unsigned *
projects / mesa.git / blobdiff