void r600_shader_binary_read_config(const struct radeon_shader_binary *binary,
struct r600_bytecode *bc,
uint64_t symbol_offset,
boolean *use_kill)
{
unsigned i;
const unsigned char *config =
radeon_shader_binary_config_start(binary, symbol_offset);
for (i = 0; i < binary->config_size_per_symbol; i+= 8) {
unsigned reg =
util_le32_to_cpu(*(uint32_t*)(config + i));
unsigned value =
util_le32_to_cpu(*(uint32_t*)(config + i + 4));
switch (reg) {
/* R600 / R700 */
case R_028850_SQ_PGM_RESOURCES_PS:
case R_028868_SQ_PGM_RESOURCES_VS:
/* Evergreen / Northern Islands */
case R_028844_SQ_PGM_RESOURCES_PS:
case R_028860_SQ_PGM_RESOURCES_VS:
case R_0288D4_SQ_PGM_RESOURCES_LS:
bc->ngpr = MAX2(bc->ngpr, G_028844_NUM_GPRS(value));
bc->nstack = MAX2(bc->nstack, G_028844_STACK_SIZE(value));
break;
case R_02880C_DB_SHADER_CONTROL:
*use_kill = G_02880C_KILL_ENABLE(value);
break;
case CM_R_0288E8_SQ_LDS_ALLOC:
bc->nlds_dw = value;
break;
}
}
}
unsigned r600_llvm_compile(
LLVMModuleRef mod,
enum radeon_family family,
struct r600_bytecode *bc,
boolean *use_kill,
unsigned dump)
{
unsigned r;
struct radeon_llvm_binary binary;
const char * gpu_family = r600_llvm_gpu_string(family);
unsigned i;
memset(&binary, 0, sizeof(struct radeon_llvm_binary));
r = radeon_llvm_compile(mod, &binary, gpu_family, dump);
assert(binary.code_size % 4 == 0);
bc->bytecode = CALLOC(1, binary.code_size);
memcpy(bc->bytecode, binary.code, binary.code_size);
bc->ndw = binary.code_size / 4;
for (i = 0; i < binary.config_size; i+= 8) {
unsigned reg =
util_le32_to_cpu(*(uint32_t*)(binary.config + i));
unsigned value =
util_le32_to_cpu(*(uint32_t*)(binary.config + i + 4));
switch (reg) {
/* R600 / R700 */
case R_028850_SQ_PGM_RESOURCES_PS:
case R_028868_SQ_PGM_RESOURCES_VS:
/* Evergreen / Northern Islands */
case R_028844_SQ_PGM_RESOURCES_PS:
case R_028860_SQ_PGM_RESOURCES_VS:
case R_0288D4_SQ_PGM_RESOURCES_LS:
bc->ngpr = G_028844_NUM_GPRS(value);
bc->nstack = G_028844_STACK_SIZE(value);
break;
case R_02880C_DB_SHADER_CONTROL:
*use_kill = G_02880C_KILL_ENABLE(value);
break;
case CM_R_0288E8_SQ_LDS_ALLOC:
bc->nlds_dw = value;
break;
}
}
FREE(binary.code);
FREE(binary.config);
return r;
}
static void si_set_global_binding(
struct pipe_context *ctx, unsigned first, unsigned n,
struct pipe_resource **resources,
uint32_t **handles)
{
unsigned i;
struct si_context *sctx = (struct si_context*)ctx;
struct si_compute *program = sctx->cs_shader_state.program;
if (!resources) {
for (i = first; i < first + n; i++) {
pipe_resource_reference(&program->global_buffers[i], NULL);
}
return;
}
for (i = first; i < first + n; i++) {
uint64_t va;
uint32_t offset;
pipe_resource_reference(&program->global_buffers[i], resources[i]);
va = r600_resource(resources[i])->gpu_address;
offset = util_le32_to_cpu(*handles[i]);
va += offset;
va = util_cpu_to_le64(va);
memcpy(handles[i], &va, sizeof(va));
}
}
static void evergreen_set_global_binding(struct pipe_context *ctx,
unsigned first, unsigned n,
struct pipe_resource **resources,
uint32_t **handles)
{
struct r600_context *rctx = (struct r600_context *)ctx;
struct compute_memory_pool *pool = rctx->screen->global_pool;
struct r600_resource_global **buffers =
(struct r600_resource_global **)resources;
unsigned i;
COMPUTE_DBG(rctx->screen, "*** evergreen_set_global_binding first = %u n = %u\n",
first, n);
if (!resources) {
/* XXX: Unset */
return;
}
/* We mark these items for promotion to the pool if they
* aren't already there */
for (i = first; i < first + n; i++) {
struct compute_memory_item *item = buffers[i]->chunk;
if (!is_item_in_pool(item))
buffers[i]->chunk->status |= ITEM_FOR_PROMOTING;
}
if (compute_memory_finalize_pending(pool, ctx) == -1) {
/* XXX: Unset */
return;
}
for (i = first; i < first + n; i++)
{
uint32_t buffer_offset;
uint32_t handle;
assert(resources[i]->target == PIPE_BUFFER);
assert(resources[i]->bind & PIPE_BIND_GLOBAL);
buffer_offset = util_le32_to_cpu(*(handles[i]));
handle = buffer_offset + buffers[i]->chunk->start_in_dw * 4;
*(handles[i]) = util_cpu_to_le32(handle);
}
/* globals for writing */
evergreen_set_rat(rctx->cs_shader_state.shader, 0, pool->bo, 0, pool->size_in_dw * 4);
/* globals for reading */
evergreen_cs_set_vertex_buffer(rctx, 1, 0,
(struct pipe_resource*)pool->bo);
/* constants for reading, LLVM puts them in text segment */
evergreen_cs_set_vertex_buffer(rctx, 2, 0,
(struct pipe_resource*)rctx->cs_shader_state.shader->code_bo);
}
static boolean r300_get_query_result(struct pipe_context* pipe,
struct pipe_query* query,
boolean wait,
union pipe_query_result *vresult)
{
struct r300_context* r300 = r300_context(pipe);
struct r300_query *q = r300_query(query);
unsigned i;
uint32_t temp, *map;
if (q->type == PIPE_QUERY_GPU_FINISHED) {
if (wait) {
r300->rws->buffer_wait(q->buf, RADEON_USAGE_READWRITE);
vresult->b = TRUE;
} else {
vresult->b = !r300->rws->buffer_is_busy(q->buf, RADEON_USAGE_READWRITE);
}
return vresult->b;
}
map = r300->rws->buffer_map(q->cs_buf, r300->cs,
PIPE_TRANSFER_READ |
(!wait ? PIPE_TRANSFER_DONTBLOCK : 0));
if (!map)
return FALSE;
/* Sum up the results. */
temp = 0;
for (i = 0; i < q->num_results; i++) {
/* Convert little endian values written by GPU to CPU byte order */
temp += util_le32_to_cpu(*map);
map++;
}
r300->rws->buffer_unmap(q->cs_buf);
if (q->type == PIPE_QUERY_OCCLUSION_PREDICATE) {
vresult->b = temp != 0;
} else {
vresult->u64 = temp;
}
return TRUE;
}
