if (!cf)
return NULL;
- LIST_INITHEAD(&cf->list);
- LIST_INITHEAD(&cf->alu);
- LIST_INITHEAD(&cf->vtx);
- LIST_INITHEAD(&cf->tex);
- LIST_INITHEAD(&cf->gds);
+ list_inithead(&cf->list);
+ list_inithead(&cf->alu);
+ list_inithead(&cf->vtx);
+ list_inithead(&cf->tex);
+ list_inithead(&cf->gds);
return cf;
}
if (!alu)
return NULL;
- LIST_INITHEAD(&alu->list);
+ list_inithead(&alu->list);
return alu;
}
if (!vtx)
return NULL;
- LIST_INITHEAD(&vtx->list);
+ list_inithead(&vtx->list);
return vtx;
}
if (!tex)
return NULL;
- LIST_INITHEAD(&tex->list);
+ list_inithead(&tex->list);
return tex;
}
if (gds == NULL)
return NULL;
- LIST_INITHEAD(&gds->list);
+ list_inithead(&gds->list);
return gds;
}
bc->r6xx_nop_after_rel_dst = 0;
}
- LIST_INITHEAD(&bc->cf);
+ list_inithead(&bc->cf);
bc->chip_class = chip_class;
bc->family = family;
bc->has_compressed_msaa_texturing = has_compressed_msaa_texturing;
if (!cf)
return -ENOMEM;
- LIST_ADDTAIL(&cf->list, &bc->cf);
+ list_addtail(&cf->list, &bc->cf);
if (bc->cf_last) {
cf->id = bc->cf_last->id + 2;
if (bc->cf_last->eg_alu_extended) {
return 0;
}
+void r600_bytecode_need_wait_ack(struct r600_bytecode *bc, boolean need_wait_ack)
+{
+ bc->need_wait_ack = need_wait_ack;
+}
+
+boolean r600_bytecode_get_need_wait_ack(struct r600_bytecode *bc)
+{
+ return bc->need_wait_ack;
+}
+
/* alu instructions that can ony exits once per group */
static int is_alu_once_inst(struct r600_bytecode_alu *alu)
{
for (i = 0; i < max_slots; ++i) {
slots[i] = result[i];
if (result[i]) {
- LIST_DEL(&result[i]->list);
+ list_del(&result[i]->list);
result[i]->last = 0;
- LIST_ADDTAIL(&result[i]->list, &bc->cf_last->alu);
+ list_addtail(&result[i]->list, &bc->cf_last->alu);
}
}
if (nalu->dst.sel >= bc->ngpr) {
bc->ngpr = nalu->dst.sel + 1;
}
- LIST_ADDTAIL(&nalu->list, &bc->cf_last->alu);
+ list_addtail(&nalu->list, &bc->cf_last->alu);
/* each alu use 2 dwords */
bc->cf_last->ndw += 2;
bc->ndw += 2;
return -EINVAL;
}
}
- LIST_ADDTAIL(&nvtx->list, &bc->cf_last->vtx);
+ list_addtail(&nvtx->list, &bc->cf_last->vtx);
/* each fetch use 4 dwords */
bc->cf_last->ndw += 4;
bc->ndw += 4;
bc->cf_last->op == CF_OP_TEX) {
struct r600_bytecode_tex *ttex;
LIST_FOR_EACH_ENTRY(ttex, &bc->cf_last->tex, list) {
- if (ttex->dst_gpr == ntex->src_gpr) {
+ if (ttex->dst_gpr == ntex->src_gpr &&
+ (ttex->dst_sel_x < 4 || ttex->dst_sel_y < 4 ||
+ ttex->dst_sel_z < 4 || ttex->dst_sel_w < 4)) {
bc->force_add_cf = 1;
break;
}
if (ntex->dst_gpr >= bc->ngpr) {
bc->ngpr = ntex->dst_gpr + 1;
}
- LIST_ADDTAIL(&ntex->list, &bc->cf_last->tex);
+ list_addtail(&ntex->list, &bc->cf_last->tex);
/* each texture fetch use 4 dwords */
bc->cf_last->ndw += 4;
bc->ndw += 4;
bc->cf_last->op = CF_OP_GDS;
}
- LIST_ADDTAIL(&ngds->list, &bc->cf_last->gds);
+ list_addtail(&ngds->list, &bc->cf_last->gds);
bc->cf_last->ndw += 4; /* each GDS uses 4 dwords */
if ((bc->cf_last->ndw / 4) >= r600_bytecode_num_tex_and_vtx_instructions(bc))
bc->force_add_cf = 1;
int r600_bytecode_add_cfinst(struct r600_bytecode *bc, unsigned op)
{
int r;
+
+ /* Emit WAIT_ACK before control flow to ensure pending writes are always acked. */
+ if (op != CF_OP_MEM_SCRATCH && bc->need_wait_ack) {
+ bc->need_wait_ack = false;
+ r = r600_bytecode_add_cfinst(bc, CF_OP_WAIT_ACK);
+ }
+
r = r600_bytecode_add_cf(bc);
if (r)
return r;
free(alu);
}
- LIST_INITHEAD(&cf->alu);
+ list_inithead(&cf->alu);
LIST_FOR_EACH_ENTRY_SAFE(tex, next_tex, &cf->tex, list) {
free(tex);
}
- LIST_INITHEAD(&cf->tex);
+ list_inithead(&cf->tex);
LIST_FOR_EACH_ENTRY_SAFE(vtx, next_vtx, &cf->vtx, list) {
free(vtx);
}
- LIST_INITHEAD(&cf->vtx);
+ list_inithead(&cf->vtx);
LIST_FOR_EACH_ENTRY_SAFE(gds, next_gds, &cf->gds, list) {
free(gds);
}
- LIST_INITHEAD(&cf->gds);
+ list_inithead(&cf->gds);
free(cf);
}
- LIST_INITHEAD(&cf->list);
+ list_inithead(&cf->list);
}
static int print_swizzle(unsigned swz)
o += fprintf(stderr, "LDS_OQ_B_POP");
need_chan = 1;
break;
+ case EG_V_SQ_ALU_SRC_TIME_LO:
+ o += fprintf(stderr, "TIME_LO");
+ break;
+ case EG_V_SQ_ALU_SRC_TIME_HI:
+ o += fprintf(stderr, "TIME_HI");
+ break;
case EG_V_SQ_ALU_SRC_SE_ID:
o += fprintf(stderr, "SE_ID");
break;
return;
}
+ if (pformat == PIPE_FORMAT_A1B5G5R5_UNORM) {
+ *format = FMT_5_5_5_1;
+ return;
+ }
+
desc = util_format_description(pformat);
if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) {
goto out_unknown;
/* Signed ints */
case UTIL_FORMAT_TYPE_SIGNED:
switch (desc->channel[i].size) {
+ case 4:
+ switch (desc->nr_channels) {
+ case 2:
+ *format = FMT_4_4;
+ break;
+ case 4:
+ *format = FMT_4_4_4_4;
+ break;
+ }
+ break;
case 8:
switch (desc->nr_channels) {
case 1:
return NULL;
}
- bytecode = r600_buffer_map_sync_with_rings(&rctx->b, shader->buffer, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_UNSYNCHRONIZED);
+ bytecode = r600_buffer_map_sync_with_rings
+ (&rctx->b, shader->buffer,
+ PIPE_TRANSFER_WRITE | PIPE_TRANSFER_UNSYNCHRONIZED | RADEON_TRANSFER_TEMPORARY);
bytecode += shader->offset / 4;
if (R600_BIG_ENDIAN) {