void
dump_assembly(void *assembly, int num_annotations, struct annotation *annotation,
struct brw_context *brw, const struct gl_program *prog)
{
const char *last_annotation_string = NULL;
const void *last_annotation_ir = NULL;
for (int i = 0; i < num_annotations; i++) {
int start_offset = annotation[i].offset;
int end_offset = annotation[i + 1].offset;
if (annotation[i].block_start) {
fprintf(stderr, " START B%d", annotation[i].block_start->num);
foreach_list_typed(struct bblock_link, predecessor_link, link,
&annotation[i].block_start->parents) {
struct bblock_t *predecessor_block = predecessor_link->block;
fprintf(stderr, " <-B%d", predecessor_block->num);
}
fprintf(stderr, "\n");
}
if (last_annotation_ir != annotation[i].ir) {
last_annotation_ir = annotation[i].ir;
if (last_annotation_ir) {
fprintf(stderr, " ");
if (!prog->Instructions)
fprint_ir(stderr, annotation[i].ir);
else {
const struct prog_instruction *pi =
(const struct prog_instruction *)annotation[i].ir;
fprintf(stderr, "%d: ",
(int)(pi - prog->Instructions));
_mesa_fprint_instruction_opt(stderr,
pi,
0, PROG_PRINT_DEBUG, NULL);
}
fprintf(stderr, "\n");
}
}
if (last_annotation_string != annotation[i].annotation) {
last_annotation_string = annotation[i].annotation;
if (last_annotation_string)
fprintf(stderr, " %s\n", last_annotation_string);
}
brw_disassemble(brw, assembly, start_offset, end_offset, stderr);
if (annotation[i].block_end) {
fprintf(stderr, " END B%d", annotation[i].block_end->num);
foreach_list_typed(struct bblock_link, successor_link, link,
&annotation[i].block_end->children) {
struct bblock_t *successor_block = successor_link->block;
fprintf(stderr, " ->B%d", successor_block->num);
}
fprintf(stderr, "\n");
}
void
dump_assembly(void *assembly, int num_annotations, struct annotation *annotation,
const struct brw_device_info *devinfo)
{
const char *last_annotation_string = NULL;
const void *last_annotation_ir = NULL;
for (int i = 0; i < num_annotations; i++) {
int start_offset = annotation[i].offset;
int end_offset = annotation[i + 1].offset;
if (annotation[i].block_start) {
fprintf(stderr, " START B%d", annotation[i].block_start->num);
foreach_list_typed(struct bblock_link, predecessor_link, link,
&annotation[i].block_start->parents) {
struct bblock_t *predecessor_block = predecessor_link->block;
fprintf(stderr, " <-B%d", predecessor_block->num);
}
fprintf(stderr, "\n");
}
if (last_annotation_ir != annotation[i].ir) {
last_annotation_ir = annotation[i].ir;
if (last_annotation_ir) {
fprintf(stderr, " ");
nir_print_instr(annotation[i].ir, stderr);
fprintf(stderr, "\n");
}
}
if (last_annotation_string != annotation[i].annotation) {
last_annotation_string = annotation[i].annotation;
if (last_annotation_string)
fprintf(stderr, " %s\n", last_annotation_string);
}
brw_disassemble(devinfo, assembly, start_offset, end_offset, stderr);
if (annotation[i].error) {
fputs(annotation[i].error, stderr);
}
if (annotation[i].block_end) {
fprintf(stderr, " END B%d", annotation[i].block_end->num);
foreach_list_typed(struct bblock_link, successor_link, link,
&annotation[i].block_end->children) {
struct bblock_t *successor_block = successor_link->block;
fprintf(stderr, " ->B%d", successor_block->num);
}
fprintf(stderr, "\n");
}
void
brw_codegen_ff_gs_prog(struct brw_context *brw,
struct brw_ff_gs_prog_key *key)
{
struct brw_ff_gs_compile c;
const GLuint *program;
void *mem_ctx;
GLuint program_size;
memset(&c, 0, sizeof(c));
c.key = *key;
c.vue_map = brw->vs.prog_data->base.vue_map;
c.nr_regs = (c.vue_map.num_slots + 1)/2;
mem_ctx = ralloc_context(NULL);
/* Begin the compilation:
*/
brw_init_codegen(brw->intelScreen->devinfo, &c.func, mem_ctx);
c.func.single_program_flow = 1;
/* For some reason the thread is spawned with only 4 channels
* unmasked.
*/
brw_set_default_mask_control(&c.func, BRW_MASK_DISABLE);
if (brw->gen >= 6) {
unsigned num_verts;
bool check_edge_flag;
/* On Sandybridge, we use the GS for implementing transform feedback
* (called "Stream Out" in the PRM).
*/
switch (key->primitive) {
case _3DPRIM_POINTLIST:
num_verts = 1;
check_edge_flag = false;
break;
case _3DPRIM_LINELIST:
case _3DPRIM_LINESTRIP:
case _3DPRIM_LINELOOP:
num_verts = 2;
check_edge_flag = false;
break;
case _3DPRIM_TRILIST:
case _3DPRIM_TRIFAN:
case _3DPRIM_TRISTRIP:
case _3DPRIM_RECTLIST:
num_verts = 3;
check_edge_flag = false;
break;
case _3DPRIM_QUADLIST:
case _3DPRIM_QUADSTRIP:
case _3DPRIM_POLYGON:
num_verts = 3;
check_edge_flag = true;
break;
default:
unreachable("Unexpected primitive type in Gen6 SOL program.");
}
gen6_sol_program(&c, key, num_verts, check_edge_flag);
} else {
/* On Gen4-5, we use the GS to decompose certain types of primitives.
* Note that primitives which don't require a GS program have already
* been weeded out by now.
*/
switch (key->primitive) {
case _3DPRIM_QUADLIST:
brw_ff_gs_quads( &c, key );
break;
case _3DPRIM_QUADSTRIP:
brw_ff_gs_quad_strip( &c, key );
break;
case _3DPRIM_LINELOOP:
brw_ff_gs_lines( &c );
break;
default:
ralloc_free(mem_ctx);
return;
}
}
brw_compact_instructions(&c.func, 0, 0, NULL);
/* get the program
*/
program = brw_get_program(&c.func, &program_size);
if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
fprintf(stderr, "gs:\n");
brw_disassemble(brw->intelScreen->devinfo, c.func.store,
0, program_size, stderr);
fprintf(stderr, "\n");
}
brw_upload_cache(&brw->cache, BRW_CACHE_FF_GS_PROG,
&c.key, sizeof(c.key),
program, program_size,
&c.prog_data, sizeof(c.prog_data),
&brw->ff_gs.prog_offset, &brw->ff_gs.prog_data);
ralloc_free(mem_ctx);
}
static void compile_clip_prog( struct brw_context *brw,
struct brw_clip_prog_key *key )
{
struct brw_clip_compile c;
const GLuint *program;
void *mem_ctx;
GLuint program_size;
memset(&c, 0, sizeof(c));
mem_ctx = ralloc_context(NULL);
/* Begin the compilation:
*/
brw_init_codegen(&brw->screen->devinfo, &c.func, mem_ctx);
c.func.single_program_flow = 1;
c.key = *key;
c.vue_map = brw->vue_map_geom_out;
/* nr_regs is the number of registers filled by reading data from the VUE.
* This program accesses the entire VUE, so nr_regs needs to be the size of
* the VUE (measured in pairs, since two slots are stored in each
* register).
*/
c.nr_regs = (c.vue_map.num_slots + 1)/2;
c.prog_data.clip_mode = c.key.clip_mode; /* XXX */
/* For some reason the thread is spawned with only 4 channels
* unmasked.
*/
brw_set_default_mask_control(&c.func, BRW_MASK_DISABLE);
/* Would ideally have the option of producing a program which could
* do all three:
*/
switch (key->primitive) {
case GL_TRIANGLES:
if (key->do_unfilled)
brw_emit_unfilled_clip( &c );
else
brw_emit_tri_clip( &c );
break;
case GL_LINES:
brw_emit_line_clip( &c );
break;
case GL_POINTS:
brw_emit_point_clip( &c );
break;
default:
unreachable("not reached");
}
brw_compact_instructions(&c.func, 0, 0, NULL);
/* get the program
*/
program = brw_get_program(&c.func, &program_size);
if (unlikely(INTEL_DEBUG & DEBUG_CLIP)) {
fprintf(stderr, "clip:\n");
brw_disassemble(&brw->screen->devinfo, c.func.store,
0, program_size, stderr);
fprintf(stderr, "\n");
}
brw_upload_cache(&brw->cache,
BRW_CACHE_CLIP_PROG,
&c.key, sizeof(c.key),
program, program_size,
&c.prog_data, sizeof(c.prog_data),
&brw->clip.prog_offset, &brw->clip.prog_data);
ralloc_free(mem_ctx);
}
const GLuint *
brw_blorp_const_color_program::compile(struct brw_context *brw,
GLuint *program_size)
{
/* Set up prog_data */
memset(&prog_data, 0, sizeof(prog_data));
prog_data.persample_msaa_dispatch = false;
alloc_regs();
brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
struct brw_reg mrf_rt_write =
retype(vec16(brw_message_reg(base_mrf)), BRW_REGISTER_TYPE_F);
uint32_t mlen, msg_type;
if (key->use_simd16_replicated_data) {
/* The message payload is a single register with the low 4 floats/ints
* filled with the constant clear color.
*/
brw_set_mask_control(&func, BRW_MASK_DISABLE);
brw_MOV(&func, vec4(brw_message_reg(base_mrf)), clear_rgba);
brw_set_mask_control(&func, BRW_MASK_ENABLE);
msg_type = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE_REPLICATED;
mlen = 1;
} else {
for (int i = 0; i < 4; i++) {
/* The message payload is pairs of registers for 16 pixels each of r,
* g, b, and a.
*/
brw_set_compression_control(&func, BRW_COMPRESSION_COMPRESSED);
brw_MOV(&func,
brw_message_reg(base_mrf + i * 2),
brw_vec1_grf(clear_rgba.nr, i));
brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
}
msg_type = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE;
mlen = 8;
}
/* Now write to the render target and terminate the thread */
brw_fb_WRITE(&func,
16 /* dispatch_width */,
base_mrf /* msg_reg_nr */,
mrf_rt_write /* src0 */,
msg_type,
BRW_BLORP_RENDERBUFFER_BINDING_TABLE_INDEX,
mlen,
0 /* response_length */,
true /* eot */,
false /* header present */);
if (unlikely(INTEL_DEBUG & DEBUG_BLORP)) {
fprintf(stderr, "Native code for BLORP clear:\n");
brw_disassemble(brw, &func.store, 0, func.next_insn_offset, stderr);
fprintf(stderr, "\n");
}
brw_compact_instructions(&func);
return brw_get_program(&func, program_size);
}
static void compile_sf_prog( struct brw_context *brw,
struct brw_sf_prog_key *key )
{
struct brw_sf_compile c;
const GLuint *program;
void *mem_ctx;
GLuint program_size;
memset(&c, 0, sizeof(c));
mem_ctx = ralloc_context(NULL);
/* Begin the compilation:
*/
brw_init_codegen(brw->intelScreen->devinfo, &c.func, mem_ctx);
c.key = *key;
c.vue_map = brw->vue_map_geom_out;
if (c.key.do_point_coord) {
/*
* gl_PointCoord is a FS instead of VS builtin variable, thus it's
* not included in c.vue_map generated in VS stage. Here we add
* it manually to let SF shader generate the needed interpolation
* coefficient for FS shader.
*/
c.vue_map.varying_to_slot[BRW_VARYING_SLOT_PNTC] = c.vue_map.num_slots;
c.vue_map.slot_to_varying[c.vue_map.num_slots++] = BRW_VARYING_SLOT_PNTC;
}
c.urb_entry_read_offset = BRW_SF_URB_ENTRY_READ_OFFSET;
c.nr_attr_regs = (c.vue_map.num_slots + 1)/2 - c.urb_entry_read_offset;
c.nr_setup_regs = c.nr_attr_regs;
c.prog_data.urb_read_length = c.nr_attr_regs;
c.prog_data.urb_entry_size = c.nr_setup_regs * 2;
c.has_flat_shading = brw_any_flat_varyings(&key->interpolation_mode);
/* Which primitive? Or all three?
*/
switch (key->primitive) {
case SF_TRIANGLES:
c.nr_verts = 3;
brw_emit_tri_setup( &c, true );
break;
case SF_LINES:
c.nr_verts = 2;
brw_emit_line_setup( &c, true );
break;
case SF_POINTS:
c.nr_verts = 1;
if (key->do_point_sprite)
brw_emit_point_sprite_setup( &c, true );
else
brw_emit_point_setup( &c, true );
break;
case SF_UNFILLED_TRIS:
c.nr_verts = 3;
brw_emit_anyprim_setup( &c );
break;
default:
unreachable("not reached");
}
/* FINISHME: SF programs use calculated jumps (i.e., JMPI with a register
* source). Compacting would be difficult.
*/
/* brw_compact_instructions(&c.func, 0, 0, NULL); */
/* get the program
*/
program = brw_get_program(&c.func, &program_size);
if (unlikely(INTEL_DEBUG & DEBUG_SF)) {
fprintf(stderr, "sf:\n");
brw_disassemble(brw->intelScreen->devinfo,
c.func.store, 0, program_size, stderr);
fprintf(stderr, "\n");
}
brw_upload_cache(&brw->cache, BRW_CACHE_SF_PROG,
&c.key, sizeof(c.key),
program, program_size,
&c.prog_data, sizeof(c.prog_data),
&brw->sf.prog_offset, &brw->sf.prog_data);
ralloc_free(mem_ctx);
}