#include "r600_sq.h"
#include "r600_opcodes.h"
#include "r600_formats.h"
+#include "r600_shader.h"
#include "r600d.h"
#include <errno.h>
return 3;
switch (bc->chip_class) {
+ default:
case R600:
case R700:
switch (alu->inst) {
case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_INT:
case EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_UINT:
return 1;
- default: R600_ERR(
- "Need instruction operand number for 0x%x.\n", alu->inst);
+ default:
+ R600_ERR("Need instruction operand number for 0x%x.\n", alu->inst);
}
break;
}
return tex;
}
-void r600_bytecode_init(struct r600_bytecode *bc, enum chip_class chip_class, enum radeon_family family)
+void r600_bytecode_init(struct r600_bytecode *bc,
+ enum chip_class chip_class,
+ enum radeon_family family,
+ enum r600_msaa_texture_mode msaa_texture_mode)
{
if ((chip_class == R600) &&
(family != CHIP_RV670 && family != CHIP_RS780 && family != CHIP_RS880)) {
LIST_INITHEAD(&bc->cf);
bc->chip_class = chip_class;
+ bc->msaa_texture_mode = msaa_texture_mode;
}
static int r600_bytecode_add_cf(struct r600_bytecode *bc)
break;
default:
R600_ERR("Unknown chip class %d.\n", bc->chip_class);
+ free(nvtx);
return -EINVAL;
}
}
static int r600_bytecode_tex_build(struct r600_bytecode *bc, struct r600_bytecode_tex *tex, unsigned id)
{
bc->bytecode[id++] = S_SQ_TEX_WORD0_TEX_INST(tex->inst) |
+ EG_S_SQ_TEX_WORD0_INST_MOD(tex->inst_mod) |
S_SQ_TEX_WORD0_RESOURCE_ID(tex->resource_id) |
S_SQ_TEX_WORD0_SRC_GPR(tex->src_gpr) |
S_SQ_TEX_WORD0_SRC_REL(tex->src_rel);
fprintf(stderr, "--------------------------------------\n");
}
-static void r600_vertex_data_type(enum pipe_format pformat,
+void r600_vertex_data_type(enum pipe_format pformat,
unsigned *format,
unsigned *num_format, unsigned *format_comp, unsigned *endian)
{
R600_ERR("unsupported vertex format %s\n", util_format_name(pformat));
}
-int r600_vertex_elements_build_fetch_shader(struct r600_context *rctx, struct r600_vertex_element *ve)
+void *r600_create_vertex_fetch_shader(struct pipe_context *ctx,
+ unsigned count,
+ const struct pipe_vertex_element *elements)
{
+ struct r600_context *rctx = (struct r600_context *)ctx;
static int dump_shaders = -1;
-
struct r600_bytecode bc;
struct r600_bytecode_vtx vtx;
- struct pipe_vertex_element *elements = ve->elements;
const struct util_format_description *desc;
unsigned fetch_resource_start = rctx->chip_class >= EVERGREEN ? 0 : 160;
unsigned format, num_format, format_comp, endian;
uint32_t *bytecode;
- int i, r;
+ int i, j, r, fs_size;
+ struct r600_fetch_shader *shader;
+
+ assert(count < 32);
memset(&bc, 0, sizeof(bc));
- r600_bytecode_init(&bc, rctx->chip_class, rctx->family);
+ r600_bytecode_init(&bc, rctx->chip_class, rctx->family,
+ rctx->screen->msaa_texture_support);
- for (i = 0; i < ve->count; i++) {
+ for (i = 0; i < count; i++) {
if (elements[i].instance_divisor > 1) {
- struct r600_bytecode_alu alu;
-
- memset(&alu, 0, sizeof(alu));
- alu.inst = BC_INST(&bc, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT);
- alu.src[0].sel = 0;
- alu.src[0].chan = 3;
-
- alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
- alu.src[1].value = (1ll << 32) / elements[i].instance_divisor + 1;
-
- alu.dst.sel = i + 1;
- alu.dst.chan = 3;
- alu.dst.write = 1;
- alu.last = 1;
-
- if ((r = r600_bytecode_add_alu(&bc, &alu))) {
- r600_bytecode_clear(&bc);
- return r;
+ if (rctx->chip_class == CAYMAN) {
+ for (j = 0; j < 4; j++) {
+ struct r600_bytecode_alu alu;
+ memset(&alu, 0, sizeof(alu));
+ alu.inst = BC_INST(&bc, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT);
+ alu.src[0].sel = 0;
+ alu.src[0].chan = 3;
+ alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
+ alu.src[1].value = (1ll << 32) / elements[i].instance_divisor + 1;
+ alu.dst.sel = i + 1;
+ alu.dst.chan = j;
+ alu.dst.write = j == 3;
+ alu.last = j == 3;
+ if ((r = r600_bytecode_add_alu(&bc, &alu))) {
+ r600_bytecode_clear(&bc);
+ return NULL;
+ }
+ }
+ } else {
+ struct r600_bytecode_alu alu;
+ memset(&alu, 0, sizeof(alu));
+ alu.inst = BC_INST(&bc, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT);
+ alu.src[0].sel = 0;
+ alu.src[0].chan = 3;
+ alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
+ alu.src[1].value = (1ll << 32) / elements[i].instance_divisor + 1;
+ alu.dst.sel = i + 1;
+ alu.dst.chan = 3;
+ alu.dst.write = 1;
+ alu.last = 1;
+ if ((r = r600_bytecode_add_alu(&bc, &alu))) {
+ r600_bytecode_clear(&bc);
+ return NULL;
+ }
}
}
}
- for (i = 0; i < ve->count; i++) {
- r600_vertex_data_type(ve->elements[i].src_format,
+ for (i = 0; i < count; i++) {
+ r600_vertex_data_type(elements[i].src_format,
&format, &num_format, &format_comp, &endian);
- desc = util_format_description(ve->elements[i].src_format);
+ desc = util_format_description(elements[i].src_format);
if (desc == NULL) {
r600_bytecode_clear(&bc);
- R600_ERR("unknown format %d\n", ve->elements[i].src_format);
- return -EINVAL;
+ R600_ERR("unknown format %d\n", elements[i].src_format);
+ return NULL;
}
if (elements[i].src_offset > 65535) {
r600_bytecode_clear(&bc);
R600_ERR("too big src_offset: %u\n", elements[i].src_offset);
- return -EINVAL;
+ return NULL;
}
memset(&vtx, 0, sizeof(vtx));
if ((r = r600_bytecode_add_vtx(&bc, &vtx))) {
r600_bytecode_clear(&bc);
- return r;
+ return NULL;
}
}
if ((r = r600_bytecode_build(&bc))) {
r600_bytecode_clear(&bc);
- return r;
+ return NULL;
}
if (dump_shaders == -1)
fprintf(stderr, "______________________________________________________________\n");
}
- ve->fs_size = bc.ndw*4;
+ fs_size = bc.ndw*4;
- ve->fetch_shader = (struct r600_resource*)
- pipe_buffer_create(rctx->context.screen,
- PIPE_BIND_CUSTOM,
- PIPE_USAGE_IMMUTABLE, ve->fs_size);
- if (ve->fetch_shader == NULL) {
+ /* Allocate the CSO. */
+ shader = CALLOC_STRUCT(r600_fetch_shader);
+ if (!shader) {
r600_bytecode_clear(&bc);
- return -ENOMEM;
+ return NULL;
}
- bytecode = rctx->ws->buffer_map(ve->fetch_shader->cs_buf, rctx->cs, PIPE_TRANSFER_WRITE);
- if (bytecode == NULL) {
+ u_suballocator_alloc(rctx->allocator_fetch_shader, fs_size, &shader->offset,
+ (struct pipe_resource**)&shader->buffer);
+ if (!shader->buffer) {
r600_bytecode_clear(&bc);
- pipe_resource_reference((struct pipe_resource**)&ve->fetch_shader, NULL);
- return -ENOMEM;
+ FREE(shader);
+ return NULL;
}
+ bytecode = rctx->ws->buffer_map(shader->buffer->cs_buf, rctx->cs,
+ PIPE_TRANSFER_WRITE | PIPE_TRANSFER_UNSYNCHRONIZED);
+ bytecode += shader->offset / 4;
+
if (R600_BIG_ENDIAN) {
- for (i = 0; i < ve->fs_size / 4; ++i) {
+ for (i = 0; i < fs_size / 4; ++i) {
bytecode[i] = bswap_32(bc.bytecode[i]);
}
} else {
- memcpy(bytecode, bc.bytecode, ve->fs_size);
+ memcpy(bytecode, bc.bytecode, fs_size);
}
+ rctx->ws->buffer_unmap(shader->buffer->cs_buf);
- rctx->ws->buffer_unmap(ve->fetch_shader->cs_buf);
r600_bytecode_clear(&bc);
-
- if (rctx->chip_class >= EVERGREEN)
- evergreen_fetch_shader(&rctx->context, ve);
- else
- r600_fetch_shader(&rctx->context, ve);
-
- return 0;
+ return shader;
}
void r600_bytecode_alu_read(struct r600_bytecode_alu *alu, uint32_t word0, uint32_t word1)
/* WORD1 */
alu->bank_swizzle = G_SQ_ALU_WORD1_BANK_SWIZZLE(word1);
+ if (alu->bank_swizzle)
+ alu->bank_swizzle_force = alu->bank_swizzle;
alu->dst.sel = G_SQ_ALU_WORD1_DST_GPR(word1);
alu->dst.rel = G_SQ_ALU_WORD1_DST_REL(word1);
alu->dst.chan = G_SQ_ALU_WORD1_DST_CHAN(word1);
G_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(word1);
}
}
+
+void r600_bytecode_export_read(struct r600_bytecode_output *output, uint32_t word0, uint32_t word1)
+{
+ output->array_base = G_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(word0);
+ output->type = G_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(word0);
+ output->gpr = G_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(word0);
+ output->elem_size = G_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(word0);
+
+ output->swizzle_x = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(word1);
+ output->swizzle_y = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(word1);
+ output->swizzle_z = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(word1);
+ output->swizzle_w = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(word1);
+ output->burst_count = G_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(word1);
+ output->end_of_program = G_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(word1);
+ output->inst = R600_S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(G_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(word1));
+ output->barrier = G_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(word1);
+}
projects / mesa.git / blobdiff