* Adam Rak <adam.rak@streamnovation.com>
*/
+#include <gelf.h>
+#include <libelf.h>
#include <stdio.h>
#include <errno.h>
-#include "ac_binary.h"
#include "pipe/p_defines.h"
#include "pipe/p_state.h"
#include "pipe/p_context.h"
#define R_028850_SQ_PGM_RESOURCES_PS 0x028850
#ifdef HAVE_OPENCL
+/*
+ * shader binary helpers.
+ */
+static void r600_shader_binary_init(struct ac_shader_binary *b)
+{
+ memset(b, 0, sizeof(*b));
+}
+
+static void r600_shader_binary_clean(struct ac_shader_binary *b)
+{
+ if (!b)
+ return;
+ FREE(b->code);
+ FREE(b->config);
+ FREE(b->rodata);
+ FREE(b->global_symbol_offsets);
+ FREE(b->relocs);
+ FREE(b->disasm_string);
+ FREE(b->llvm_ir_string);
+}
+
+static void parse_symbol_table(Elf_Data *symbol_table_data,
+ const GElf_Shdr *symbol_table_header,
+ struct ac_shader_binary *binary)
+{
+ GElf_Sym symbol;
+ unsigned i = 0;
+ unsigned symbol_count =
+ symbol_table_header->sh_size / symbol_table_header->sh_entsize;
+
+ /* We are over allocating this list, because symbol_count gives the
+ * total number of symbols, and we will only be filling the list
+ * with offsets of global symbols. The memory savings from
+ * allocating the correct size of this list will be small, and
+ * I don't think it is worth the cost of pre-computing the number
+ * of global symbols.
+ */
+ binary->global_symbol_offsets = CALLOC(symbol_count, sizeof(uint64_t));
+
+ while (gelf_getsym(symbol_table_data, i++, &symbol)) {
+ unsigned i;
+ if (GELF_ST_BIND(symbol.st_info) != STB_GLOBAL ||
+ symbol.st_shndx == 0 /* Undefined symbol */) {
+ continue;
+ }
+
+ binary->global_symbol_offsets[binary->global_symbol_count] =
+ symbol.st_value;
+
+ /* Sort the list using bubble sort. This list will usually
+ * be small. */
+ for (i = binary->global_symbol_count; i > 0; --i) {
+ uint64_t lhs = binary->global_symbol_offsets[i - 1];
+ uint64_t rhs = binary->global_symbol_offsets[i];
+ if (lhs < rhs) {
+ break;
+ }
+ binary->global_symbol_offsets[i] = lhs;
+ binary->global_symbol_offsets[i - 1] = rhs;
+ }
+ ++binary->global_symbol_count;
+ }
+}
+
+
+static void parse_relocs(Elf *elf, Elf_Data *relocs, Elf_Data *symbols,
+ unsigned symbol_sh_link,
+ struct ac_shader_binary *binary)
+{
+ unsigned i;
+
+ if (!relocs || !symbols || !binary->reloc_count) {
+ return;
+ }
+ binary->relocs = CALLOC(binary->reloc_count,
+ sizeof(struct ac_shader_reloc));
+ for (i = 0; i < binary->reloc_count; i++) {
+ GElf_Sym symbol;
+ GElf_Rel rel;
+ char *symbol_name;
+ struct ac_shader_reloc *reloc = &binary->relocs[i];
+
+ gelf_getrel(relocs, i, &rel);
+ gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &symbol);
+ symbol_name = elf_strptr(elf, symbol_sh_link, symbol.st_name);
+
+ reloc->offset = rel.r_offset;
+ strncpy(reloc->name, symbol_name, sizeof(reloc->name)-1);
+ reloc->name[sizeof(reloc->name)-1] = 0;
+ }
+}
+
+static void r600_elf_read(const char *elf_data, unsigned elf_size,
+ struct ac_shader_binary *binary)
+{
+ char *elf_buffer;
+ Elf *elf;
+ Elf_Scn *section = NULL;
+ Elf_Data *symbols = NULL, *relocs = NULL;
+ size_t section_str_index;
+ unsigned symbol_sh_link = 0;
+
+ /* One of the libelf implementations
+ * (http://www.mr511.de/software/english.htm) requires calling
+ * elf_version() before elf_memory().
+ */
+ elf_version(EV_CURRENT);
+ elf_buffer = MALLOC(elf_size);
+ memcpy(elf_buffer, elf_data, elf_size);
+
+ elf = elf_memory(elf_buffer, elf_size);
+
+ elf_getshdrstrndx(elf, §ion_str_index);
+
+ while ((section = elf_nextscn(elf, section))) {
+ const char *name;
+ Elf_Data *section_data = NULL;
+ GElf_Shdr section_header;
+ if (gelf_getshdr(section, §ion_header) != §ion_header) {
+ fprintf(stderr, "Failed to read ELF section header\n");
+ return;
+ }
+ name = elf_strptr(elf, section_str_index, section_header.sh_name);
+ if (!strcmp(name, ".text")) {
+ section_data = elf_getdata(section, section_data);
+ binary->code_size = section_data->d_size;
+ binary->code = MALLOC(binary->code_size * sizeof(unsigned char));
+ memcpy(binary->code, section_data->d_buf, binary->code_size);
+ } else if (!strcmp(name, ".AMDGPU.config")) {
+ section_data = elf_getdata(section, section_data);
+ binary->config_size = section_data->d_size;
+ binary->config = MALLOC(binary->config_size * sizeof(unsigned char));
+ memcpy(binary->config, section_data->d_buf, binary->config_size);
+ } else if (!strcmp(name, ".AMDGPU.disasm")) {
+ /* Always read disassembly if it's available. */
+ section_data = elf_getdata(section, section_data);
+ binary->disasm_string = strndup(section_data->d_buf,
+ section_data->d_size);
+ } else if (!strncmp(name, ".rodata", 7)) {
+ section_data = elf_getdata(section, section_data);
+ binary->rodata_size = section_data->d_size;
+ binary->rodata = MALLOC(binary->rodata_size * sizeof(unsigned char));
+ memcpy(binary->rodata, section_data->d_buf, binary->rodata_size);
+ } else if (!strncmp(name, ".symtab", 7)) {
+ symbols = elf_getdata(section, section_data);
+ symbol_sh_link = section_header.sh_link;
+ parse_symbol_table(symbols, §ion_header, binary);
+ } else if (!strcmp(name, ".rel.text")) {
+ relocs = elf_getdata(section, section_data);
+ binary->reloc_count = section_header.sh_size /
+ section_header.sh_entsize;
+ }
+ }
+
+ parse_relocs(elf, relocs, symbols, symbol_sh_link, binary);
+
+ if (elf){
+ elf_end(elf);
+ }
+ FREE(elf_buffer);
+
+ /* Cache the config size per symbol */
+ if (binary->global_symbol_count) {
+ binary->config_size_per_symbol =
+ binary->config_size / binary->global_symbol_count;
+ } else {
+ binary->global_symbol_count = 1;
+ binary->config_size_per_symbol = binary->config_size;
+ }
+}
+
+static const unsigned char *r600_shader_binary_config_start(
+ const struct ac_shader_binary *binary,
+ uint64_t symbol_offset)
+{
+ unsigned i;
+ for (i = 0; i < binary->global_symbol_count; ++i) {
+ if (binary->global_symbol_offsets[i] == symbol_offset) {
+ unsigned offset = i * binary->config_size_per_symbol;
+ return binary->config + offset;
+ }
+ }
+ return binary->config;
+}
static void r600_shader_binary_read_config(const struct ac_shader_binary *binary,
struct r600_bytecode *bc,
{
unsigned i;
const unsigned char *config =
- ac_shader_binary_config_start(binary, symbol_offset);
+ r600_shader_binary_config_start(binary, symbol_offset);
for (i = 0; i < binary->config_size_per_symbol; i+= 8) {
unsigned reg =
COMPUTE_DBG(rctx->screen, "*** evergreen_create_compute_state\n");
header = cso->prog;
code = cso->prog + sizeof(struct pipe_llvm_program_header);
- radeon_shader_binary_init(&shader->binary);
- ac_elf_read(code, header->num_bytes, &shader->binary);
+ r600_shader_binary_init(&shader->binary);
+ r600_elf_read(code, header->num_bytes, &shader->binary);
r600_create_shader(&shader->bc, &shader->binary, &use_kill);
/* Upload code + ROdata */
if (!shader)
return;
- radeon_shader_binary_clean(&shader->binary);
+#ifdef HAVE_OPENCL
+ r600_shader_binary_clean(&shader->binary);
+#endif
r600_destroy_shader(&shader->bc);
/* TODO destroy shader->code_bo, shader->const_bo
projects / mesa.git / blobdiff