* DAMAGES OR OTHER 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:
- * Marek Olšák <maraeo@gmail.com>
*/
#include "ac_debug.h"
#include <memcheck.h>
#define VG(x) x
#else
-#define VG(x)
+#define VG(x) ((void)0)
#endif
#include <inttypes.h>
#include "sid.h"
-#include "gfx9d.h"
#include "sid_tables.h"
#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/u_string.h"
+#include <assert.h>
+
/* Parsed IBs are difficult to read without colors. Use "less -R file" to
* read them, or use "aha -b -f file" to convert them to html.
*/
print_value(file, value, bits);
}
-void ac_dump_reg(FILE *file, unsigned offset, uint32_t value,
- uint32_t field_mask)
+static const struct si_reg *find_register(enum chip_class chip_class, unsigned offset)
{
- int r, f;
-
- for (r = 0; r < ARRAY_SIZE(sid_reg_table); r++) {
- const struct si_reg *reg = &sid_reg_table[r];
- const char *reg_name = sid_strings + reg->name_offset;
+ const struct si_reg *table;
+ unsigned table_size;
+
+ switch (chip_class) {
+ case GFX10_3:
+ case GFX10:
+ table = gfx10_reg_table;
+ table_size = ARRAY_SIZE(gfx10_reg_table);
+ break;
+ case GFX9:
+ table = gfx9_reg_table;
+ table_size = ARRAY_SIZE(gfx9_reg_table);
+ break;
+ case GFX8:
+ table = gfx8_reg_table;
+ table_size = ARRAY_SIZE(gfx8_reg_table);
+ break;
+ case GFX7:
+ table = gfx7_reg_table;
+ table_size = ARRAY_SIZE(gfx7_reg_table);
+ break;
+ case GFX6:
+ table = gfx6_reg_table;
+ table_size = ARRAY_SIZE(gfx6_reg_table);
+ break;
+ default:
+ return NULL;
+ }
- if (reg->offset == offset) {
- bool first_field = true;
+ for (unsigned i = 0; i < table_size; i++) {
+ const struct si_reg *reg = &table[i];
- print_spaces(file, INDENT_PKT);
- fprintf(file, COLOR_YELLOW "%s" COLOR_RESET " <- ",
- reg_name);
+ if (reg->offset == offset)
+ return reg;
+ }
- if (!reg->num_fields) {
- print_value(file, value, 32);
- return;
- }
+ return NULL;
+}
- for (f = 0; f < reg->num_fields; f++) {
- const struct si_field *field = sid_fields_table + reg->fields_offset + f;
- const int *values_offsets = sid_strings_offsets + field->values_offset;
- uint32_t val = (value & field->mask) >>
- (ffs(field->mask) - 1);
+const char *ac_get_register_name(enum chip_class chip_class, unsigned offset)
+{
+ const struct si_reg *reg = find_register(chip_class, offset);
- if (!(field->mask & field_mask))
- continue;
+ return reg ? sid_strings + reg->name_offset : "(no name)";
+}
- /* Indent the field. */
- if (!first_field)
- print_spaces(file,
- INDENT_PKT + strlen(reg_name) + 4);
+void ac_dump_reg(FILE *file, enum chip_class chip_class, unsigned offset,
+ uint32_t value, uint32_t field_mask)
+{
+ const struct si_reg *reg = find_register(chip_class, offset);
- /* Print the field. */
- fprintf(file, "%s = ", sid_strings + field->name_offset);
+ if (reg) {
+ const char *reg_name = sid_strings + reg->name_offset;
+ bool first_field = true;
- if (val < field->num_values && values_offsets[val] >= 0)
- fprintf(file, "%s\n", sid_strings + values_offsets[val]);
- else
- print_value(file, val,
- util_bitcount(field->mask));
+ print_spaces(file, INDENT_PKT);
+ fprintf(file, COLOR_YELLOW "%s" COLOR_RESET " <- ",
+ reg_name);
- first_field = false;
- }
+ if (!reg->num_fields) {
+ print_value(file, value, 32);
return;
}
+
+ for (unsigned f = 0; f < reg->num_fields; f++) {
+ const struct si_field *field = sid_fields_table + reg->fields_offset + f;
+ const int *values_offsets = sid_strings_offsets + field->values_offset;
+ uint32_t val = (value & field->mask) >>
+ (ffs(field->mask) - 1);
+
+ if (!(field->mask & field_mask))
+ continue;
+
+ /* Indent the field. */
+ if (!first_field)
+ print_spaces(file,
+ INDENT_PKT + strlen(reg_name) + 4);
+
+ /* Print the field. */
+ fprintf(file, "%s = ", sid_strings + field->name_offset);
+
+ if (val < field->num_values && values_offsets[val] >= 0)
+ fprintf(file, "%s\n", sid_strings + values_offsets[val]);
+ else
+ print_value(file, val,
+ util_bitcount(field->mask));
+
+ first_field = false;
+ }
+ return;
}
print_spaces(file, INDENT_PKT);
}
for (i = 0; i < count; i++)
- ac_dump_reg(f, reg + i*4, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, reg + i*4, ac_ib_get(ib), ~0);
}
static void ac_parse_packet3(FILE *f, uint32_t header, struct ac_ib_parser *ib,
if (op == PKT3_SET_CONTEXT_REG ||
op == PKT3_SET_CONFIG_REG ||
op == PKT3_SET_UCONFIG_REG ||
+ op == PKT3_SET_UCONFIG_REG_INDEX ||
op == PKT3_SET_SH_REG)
fprintf(f, COLOR_CYAN "%s%s" COLOR_CYAN ":\n",
name, predicate);
ac_parse_set_reg_packet(f, count, SI_CONFIG_REG_OFFSET, ib);
break;
case PKT3_SET_UCONFIG_REG:
+ case PKT3_SET_UCONFIG_REG_INDEX:
ac_parse_set_reg_packet(f, count, CIK_UCONFIG_REG_OFFSET, ib);
break;
case PKT3_SET_SH_REG:
ac_parse_set_reg_packet(f, count, SI_SH_REG_OFFSET, ib);
break;
case PKT3_ACQUIRE_MEM:
- ac_dump_reg(f, R_0301F0_CP_COHER_CNTL, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0301F4_CP_COHER_SIZE, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_030230_CP_COHER_SIZE_HI, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0301F8_CP_COHER_BASE, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0301E4_CP_COHER_BASE_HI, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0301F0_CP_COHER_CNTL, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0301F4_CP_COHER_SIZE, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_030230_CP_COHER_SIZE_HI, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0301F8_CP_COHER_BASE, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0301E4_CP_COHER_BASE_HI, ac_ib_get(ib), ~0);
print_named_value(f, "POLL_INTERVAL", ac_ib_get(ib), 16);
+ if (ib->chip_class >= GFX10)
+ ac_dump_reg(f, ib->chip_class, R_586_GCR_CNTL, ac_ib_get(ib), ~0);
break;
case PKT3_SURFACE_SYNC:
- if (ib->chip_class >= CIK) {
- ac_dump_reg(f, R_0301F0_CP_COHER_CNTL, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0301F4_CP_COHER_SIZE, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0301F8_CP_COHER_BASE, ac_ib_get(ib), ~0);
+ if (ib->chip_class >= GFX7) {
+ ac_dump_reg(f, ib->chip_class, R_0301F0_CP_COHER_CNTL, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0301F4_CP_COHER_SIZE, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0301F8_CP_COHER_BASE, ac_ib_get(ib), ~0);
} else {
- ac_dump_reg(f, R_0085F0_CP_COHER_CNTL, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0085F4_CP_COHER_SIZE, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0085F8_CP_COHER_BASE, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0085F0_CP_COHER_CNTL, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0085F4_CP_COHER_SIZE, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0085F8_CP_COHER_BASE, ac_ib_get(ib), ~0);
}
print_named_value(f, "POLL_INTERVAL", ac_ib_get(ib), 16);
break;
case PKT3_EVENT_WRITE: {
uint32_t event_dw = ac_ib_get(ib);
- ac_dump_reg(f, R_028A90_VGT_EVENT_INITIATOR, event_dw,
+ ac_dump_reg(f, ib->chip_class, R_028A90_VGT_EVENT_INITIATOR, event_dw,
S_028A90_EVENT_TYPE(~0));
print_named_value(f, "EVENT_INDEX", (event_dw >> 8) & 0xf, 4);
print_named_value(f, "INV_L2", (event_dw >> 20) & 0x1, 1);
}
case PKT3_EVENT_WRITE_EOP: {
uint32_t event_dw = ac_ib_get(ib);
- ac_dump_reg(f, R_028A90_VGT_EVENT_INITIATOR, event_dw,
+ ac_dump_reg(f, ib->chip_class, R_028A90_VGT_EVENT_INITIATOR, event_dw,
S_028A90_EVENT_TYPE(~0));
print_named_value(f, "EVENT_INDEX", (event_dw >> 8) & 0xf, 4);
print_named_value(f, "TCL1_VOL_ACTION_ENA", (event_dw >> 12) & 0x1, 1);
}
case PKT3_RELEASE_MEM: {
uint32_t event_dw = ac_ib_get(ib);
- ac_dump_reg(f, R_028A90_VGT_EVENT_INITIATOR, event_dw,
- S_028A90_EVENT_TYPE(~0));
- print_named_value(f, "EVENT_INDEX", (event_dw >> 8) & 0xf, 4);
- print_named_value(f, "TCL1_VOL_ACTION_ENA", (event_dw >> 12) & 0x1, 1);
- print_named_value(f, "TC_VOL_ACTION_ENA", (event_dw >> 13) & 0x1, 1);
- print_named_value(f, "TC_WB_ACTION_ENA", (event_dw >> 15) & 0x1, 1);
- print_named_value(f, "TCL1_ACTION_ENA", (event_dw >> 16) & 0x1, 1);
- print_named_value(f, "TC_ACTION_ENA", (event_dw >> 17) & 0x1, 1);
- print_named_value(f, "TC_NC_ACTION_ENA", (event_dw >> 19) & 0x1, 1);
- print_named_value(f, "TC_WC_ACTION_ENA", (event_dw >> 20) & 0x1, 1);
- print_named_value(f, "TC_MD_ACTION_ENA", (event_dw >> 21) & 0x1, 1);
+ if (ib->chip_class >= GFX10) {
+ ac_dump_reg(f, ib->chip_class, R_490_RELEASE_MEM_OP, event_dw, ~0u);
+ } else {
+ ac_dump_reg(f, ib->chip_class, R_028A90_VGT_EVENT_INITIATOR, event_dw,
+ S_028A90_EVENT_TYPE(~0));
+ print_named_value(f, "EVENT_INDEX", (event_dw >> 8) & 0xf, 4);
+ print_named_value(f, "TCL1_VOL_ACTION_ENA", (event_dw >> 12) & 0x1, 1);
+ print_named_value(f, "TC_VOL_ACTION_ENA", (event_dw >> 13) & 0x1, 1);
+ print_named_value(f, "TC_WB_ACTION_ENA", (event_dw >> 15) & 0x1, 1);
+ print_named_value(f, "TCL1_ACTION_ENA", (event_dw >> 16) & 0x1, 1);
+ print_named_value(f, "TC_ACTION_ENA", (event_dw >> 17) & 0x1, 1);
+ print_named_value(f, "TC_NC_ACTION_ENA", (event_dw >> 19) & 0x1, 1);
+ print_named_value(f, "TC_WC_ACTION_ENA", (event_dw >> 20) & 0x1, 1);
+ print_named_value(f, "TC_MD_ACTION_ENA", (event_dw >> 21) & 0x1, 1);
+ }
uint32_t sel_dw = ac_ib_get(ib);
print_named_value(f, "DST_SEL", (sel_dw >> 16) & 0x3, 2);
print_named_value(f, "INT_SEL", (sel_dw >> 24) & 0x7, 3);
print_named_value(f, "POLL_INTERVAL", ac_ib_get(ib), 16);
break;
case PKT3_DRAW_INDEX_AUTO:
- ac_dump_reg(f, R_030930_VGT_NUM_INDICES, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0287F0_VGT_DRAW_INITIATOR, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_030930_VGT_NUM_INDICES, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0287F0_VGT_DRAW_INITIATOR, ac_ib_get(ib), ~0);
break;
case PKT3_DRAW_INDEX_2:
- ac_dump_reg(f, R_028A78_VGT_DMA_MAX_SIZE, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0287E8_VGT_DMA_BASE, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0287E4_VGT_DMA_BASE_HI, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_030930_VGT_NUM_INDICES, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_0287F0_VGT_DRAW_INITIATOR, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_028A78_VGT_DMA_MAX_SIZE, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0287E8_VGT_DMA_BASE, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0287E4_VGT_DMA_BASE_HI, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_030930_VGT_NUM_INDICES, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_0287F0_VGT_DRAW_INITIATOR, ac_ib_get(ib), ~0);
break;
case PKT3_INDEX_TYPE:
- ac_dump_reg(f, R_028A7C_VGT_DMA_INDEX_TYPE, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_028A7C_VGT_DMA_INDEX_TYPE, ac_ib_get(ib), ~0);
break;
case PKT3_NUM_INSTANCES:
- ac_dump_reg(f, R_030934_VGT_NUM_INSTANCES, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_030934_VGT_NUM_INSTANCES, ac_ib_get(ib), ~0);
break;
case PKT3_WRITE_DATA:
- ac_dump_reg(f, R_370_CONTROL, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_371_DST_ADDR_LO, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_372_DST_ADDR_HI, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_370_CONTROL, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_371_DST_ADDR_LO, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_372_DST_ADDR_HI, ac_ib_get(ib), ~0);
/* The payload is written automatically */
break;
case PKT3_CP_DMA:
- ac_dump_reg(f, R_410_CP_DMA_WORD0, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_411_CP_DMA_WORD1, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_412_CP_DMA_WORD2, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_413_CP_DMA_WORD3, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_414_COMMAND, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_410_CP_DMA_WORD0, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_411_CP_DMA_WORD1, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_412_CP_DMA_WORD2, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_413_CP_DMA_WORD3, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_414_COMMAND, ac_ib_get(ib), ~0);
break;
case PKT3_DMA_DATA:
- ac_dump_reg(f, R_500_DMA_DATA_WORD0, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_501_SRC_ADDR_LO, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_502_SRC_ADDR_HI, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_503_DST_ADDR_LO, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_504_DST_ADDR_HI, ac_ib_get(ib), ~0);
- ac_dump_reg(f, R_414_COMMAND, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_500_DMA_DATA_WORD0, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_501_SRC_ADDR_LO, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_502_SRC_ADDR_HI, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_503_DST_ADDR_LO, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_504_DST_ADDR_HI, ac_ib_get(ib), ~0);
+ ac_dump_reg(f, ib->chip_class, R_414_COMMAND, ac_ib_get(ib), ~0);
break;
case PKT3_INDIRECT_BUFFER_SI:
case PKT3_INDIRECT_BUFFER_CONST:
case PKT3_INDIRECT_BUFFER_CIK: {
uint32_t base_lo_dw = ac_ib_get(ib);
- ac_dump_reg(f, R_3F0_IB_BASE_LO, base_lo_dw, ~0);
+ ac_dump_reg(f, ib->chip_class, R_3F0_IB_BASE_LO, base_lo_dw, ~0);
uint32_t base_hi_dw = ac_ib_get(ib);
- ac_dump_reg(f, R_3F1_IB_BASE_HI, base_hi_dw, ~0);
+ ac_dump_reg(f, ib->chip_class, R_3F1_IB_BASE_HI, base_hi_dw, ~0);
uint32_t control_dw = ac_ib_get(ib);
- ac_dump_reg(f, R_3F2_CONTROL, control_dw, ~0);
+ ac_dump_reg(f, ib->chip_class, R_3F2_IB_CONTROL, control_dw, ~0);
if (!ib->addr_callback)
break;
case PKT3_PFP_SYNC_ME:
break;
case PKT3_NOP:
- if (header == 0xffff1000) {
+ if (header == PKT3_NOP_PAD) {
count = -1; /* One dword NOP. */
} else if (count == 0 && ib->cur_dw < ib->num_dw &&
AC_IS_TRACE_POINT(ib->ib[ib->cur_dw])) {
if (indent)
print_spaces(f, indent);
- char *end = util_strchrnul(out, '\n');
+ char *end = strchrnul(out, '\n');
fwrite(out, end - out, 1, f);
fputc('\n', f); /* always end with a new line */
if (!*end)
return fault;
}
+
+static int compare_wave(const void *p1, const void *p2)
+{
+ struct ac_wave_info *w1 = (struct ac_wave_info *)p1;
+ struct ac_wave_info *w2 = (struct ac_wave_info *)p2;
+
+ /* Sort waves according to PC and then SE, SH, CU, etc. */
+ if (w1->pc < w2->pc)
+ return -1;
+ if (w1->pc > w2->pc)
+ return 1;
+ if (w1->se < w2->se)
+ return -1;
+ if (w1->se > w2->se)
+ return 1;
+ if (w1->sh < w2->sh)
+ return -1;
+ if (w1->sh > w2->sh)
+ return 1;
+ if (w1->cu < w2->cu)
+ return -1;
+ if (w1->cu > w2->cu)
+ return 1;
+ if (w1->simd < w2->simd)
+ return -1;
+ if (w1->simd > w2->simd)
+ return 1;
+ if (w1->wave < w2->wave)
+ return -1;
+ if (w1->wave > w2->wave)
+ return 1;
+
+ return 0;
+}
+
+/* Return wave information. "waves" should be a large enough array. */
+unsigned ac_get_wave_info(enum chip_class chip_class,
+ struct ac_wave_info waves[AC_MAX_WAVES_PER_CHIP])
+{
+ char line[2000], cmd[128];
+ unsigned num_waves = 0;
+
+ sprintf(cmd, "umr -O halt_waves -wa %s", chip_class >= GFX10 ? "gfx_0.0.0" : "gfx");
+
+ FILE *p = popen(cmd, "r");
+ if (!p)
+ return 0;
+
+ if (!fgets(line, sizeof(line), p) ||
+ strncmp(line, "SE", 2) != 0) {
+ pclose(p);
+ return 0;
+ }
+
+ while (fgets(line, sizeof(line), p)) {
+ struct ac_wave_info *w;
+ uint32_t pc_hi, pc_lo, exec_hi, exec_lo;
+
+ assert(num_waves < AC_MAX_WAVES_PER_CHIP);
+ w = &waves[num_waves];
+
+ if (sscanf(line, "%u %u %u %u %u %x %x %x %x %x %x %x",
+ &w->se, &w->sh, &w->cu, &w->simd, &w->wave,
+ &w->status, &pc_hi, &pc_lo, &w->inst_dw0,
+ &w->inst_dw1, &exec_hi, &exec_lo) == 12) {
+ w->pc = ((uint64_t)pc_hi << 32) | pc_lo;
+ w->exec = ((uint64_t)exec_hi << 32) | exec_lo;
+ w->matched = false;
+ num_waves++;
+ }
+ }
+
+ qsort(waves, num_waves, sizeof(struct ac_wave_info), compare_wave);
+
+ pclose(p);
+ return num_waves;
+}