* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
- *
- * Authors: Tom Stellard <thomas.stellard@amd.com>
- *
- * Based on radeon_elf_util.c.
*/
#include "ac_binary.h"
}
}
-void ac_elf_read(const char *elf_data, unsigned elf_size,
+bool ac_elf_read(const char *elf_data, unsigned elf_size,
struct ac_shader_binary *binary)
{
char *elf_buffer;
Elf_Data *symbols = NULL, *relocs = NULL;
size_t section_str_index;
unsigned symbol_sh_link = 0;
+ bool success = true;
/* One of the libelf implementations
* (http://www.mr511.de/software/english.htm) requires calling
GElf_Shdr section_header;
if (gelf_getshdr(section, §ion_header) != §ion_header) {
fprintf(stderr, "Failed to read ELF section header\n");
- return;
+ success = false;
+ break;
}
name = elf_strptr(elf, section_str_index, section_header.sh_name);
if (!strcmp(name, ".text")) {
} else if (!strcmp(name, ".AMDGPU.config")) {
section_data = elf_getdata(section, section_data);
binary->config_size = section_data->d_size;
+ if (!binary->config_size) {
+ fprintf(stderr, ".AMDGPU.config is empty!\n");
+ success = false;
+ break;
+ }
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")) {
binary->global_symbol_count = 1;
binary->config_size_per_symbol = binary->config_size;
}
+ return success;
}
-static
const unsigned char *ac_shader_binary_config_start(
const struct ac_shader_binary *binary,
uint64_t symbol_offset)
return binary->config;
}
-
-static const char *scratch_rsrc_dword0_symbol =
- "SCRATCH_RSRC_DWORD0";
-
-static const char *scratch_rsrc_dword1_symbol =
- "SCRATCH_RSRC_DWORD1";
-
-void ac_shader_binary_read_config(struct ac_shader_binary *binary,
- struct ac_shader_config *conf,
- unsigned symbol_offset)
+/* Parse configuration data in .AMDGPU.config section format. */
+void ac_parse_shader_binary_config(const char *data, size_t nbytes,
+ bool really_needs_scratch,
+ struct ac_shader_config *conf)
{
- unsigned i;
- const unsigned char *config =
- ac_shader_binary_config_start(binary, symbol_offset);
- bool really_needs_scratch = false;
-
- /* LLVM adds SGPR spills to the scratch size.
- * Find out if we really need the scratch buffer.
- */
- for (i = 0; i < binary->reloc_count; i++) {
- const struct ac_shader_reloc *reloc = &binary->relocs[i];
+ uint32_t wavesize = 0;
- if (!strcmp(scratch_rsrc_dword0_symbol, reloc->name) ||
- !strcmp(scratch_rsrc_dword1_symbol, reloc->name)) {
- really_needs_scratch = true;
- break;
- }
- }
-
- 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));
+ for (size_t i = 0; i < nbytes; i += 8) {
+ unsigned reg = util_le32_to_cpu(*(uint32_t*)(data + i));
+ unsigned value = util_le32_to_cpu(*(uint32_t*)(data + i + 4));
switch (reg) {
case R_00B028_SPI_SHADER_PGM_RSRC1_PS:
case R_00B128_SPI_SHADER_PGM_RSRC1_VS:
case R_00B228_SPI_SHADER_PGM_RSRC1_GS:
case R_00B848_COMPUTE_PGM_RSRC1:
+ case R_00B428_SPI_SHADER_PGM_RSRC1_HS:
conf->num_sgprs = MAX2(conf->num_sgprs, (G_00B028_SGPRS(value) + 1) * 8);
conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 4);
conf->float_mode = G_00B028_FLOAT_MODE(value);
+ conf->rsrc1 = value;
break;
case R_00B02C_SPI_SHADER_PGM_RSRC2_PS:
conf->lds_size = MAX2(conf->lds_size, G_00B02C_EXTRA_LDS_SIZE(value));
break;
case R_00B84C_COMPUTE_PGM_RSRC2:
conf->lds_size = MAX2(conf->lds_size, G_00B84C_LDS_SIZE(value));
+ conf->rsrc2 = value;
break;
case R_0286CC_SPI_PS_INPUT_ENA:
conf->spi_ps_input_ena = value;
case R_0286E8_SPI_TMPRING_SIZE:
case R_00B860_COMPUTE_TMPRING_SIZE:
/* WAVESIZE is in units of 256 dwords. */
- if (really_needs_scratch)
- conf->scratch_bytes_per_wave =
- G_00B860_WAVESIZE(value) * 256 * 4;
+ wavesize = value;
break;
case SPILLED_SGPRS:
conf->spilled_sgprs = value;
}
break;
}
+ }
+
+ if (!conf->spi_ps_input_addr)
+ conf->spi_ps_input_addr = conf->spi_ps_input_ena;
- if (!conf->spi_ps_input_addr)
- conf->spi_ps_input_addr = conf->spi_ps_input_ena;
+ if (really_needs_scratch) {
+ /* sgprs spills aren't spilling */
+ conf->scratch_bytes_per_wave = G_00B860_WAVESIZE(wavesize) * 256 * 4;
}
}
+
+static const char *scratch_rsrc_dword0_symbol =
+ "SCRATCH_RSRC_DWORD0";
+
+static const char *scratch_rsrc_dword1_symbol =
+ "SCRATCH_RSRC_DWORD1";
+
+void ac_shader_binary_read_config(struct ac_shader_binary *binary,
+ struct ac_shader_config *conf,
+ unsigned symbol_offset,
+ bool supports_spill)
+{
+ unsigned i;
+ const char *config =
+ (const char *)ac_shader_binary_config_start(binary, symbol_offset);
+ bool really_needs_scratch = false;
+ /* LLVM adds SGPR spills to the scratch size.
+ * Find out if we really need the scratch buffer.
+ */
+ if (supports_spill) {
+ really_needs_scratch = true;
+ } else {
+ for (i = 0; i < binary->reloc_count; i++) {
+ const struct ac_shader_reloc *reloc = &binary->relocs[i];
+
+ if (!strcmp(scratch_rsrc_dword0_symbol, reloc->name) ||
+ !strcmp(scratch_rsrc_dword1_symbol, reloc->name)) {
+ really_needs_scratch = true;
+ break;
+ }
+ }
+ }
+
+ ac_parse_shader_binary_config(config, binary->config_size_per_symbol,
+ really_needs_scratch, conf);
+}
+
+void ac_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);
+}