Emulation/cen64
1
0
Fork 0
mirror of https://github.com/n64dev/cen64.git synced 2024-05-11 17:05:33 -04:00
Code Issues Packages Projects Releases Wiki Activity

Teach the profiler about L1D misses

Browse source
This commit is contained in:
Lauri Kasanen 2020-12-07 12:23:30 +02:00
parent 9464379f8a
commit 81bf10960f
4 changed files with 26 additions and 11 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
device/device.c
View file

@ -151,10 +151,11 @@ void device_destroy(struct cen64_device *device, const char *cart_path) {
uint32_t i;
for (i = 0; i < 8 * 1024 * 1024; i++) {
uint64_t sample = get_profile_sample(device->vr4300, i);
if (sample < 10)
const uint64_t sample = get_profile_sample(device->vr4300, i);
const uint64_t l1d_sample = get_profile_sample(device->vr4300, i + (8 * 1024 * 1024));
if (sample < 10 && l1d_sample < 10)
continue;
fprintf(f, "%x %lu\n", i + 0x80000000, sample);
fprintf(f, "%x %lu %lu\n", i + 0x80000000, sample, l1d_sample);
}
fclose(f);

15
util/cen64-profile2callgrind.cpp
View file

@ -130,11 +130,11 @@ int main(int argc, char **argv) {
fprintf(out, "# callgrind format\n");
fprintf(out, "cmd: %s\n", argv[2]);
fprintf(out, "events: instructions\n\n");
fprintf(out, "events: instructions l1d-misses\n\n");
fprintf(out, "ob=%s\n\n", argv[2]);
uint64_t summary = 0;
uint64_t summary = 0, l1d_summary = 0;
// We have a list of functions, now turn the samples into C line info
f = fopen(argv[1], "r");
@ -142,11 +142,12 @@ int main(int argc, char **argv) {
while (fgets(buf, PATH_MAX, f)) {
uint32_t addr;
uint64_t num;
if (sscanf(buf, "%x %lu", &addr, &num) != 2)
uint64_t num, l1d_num;
if (sscanf(buf, "%x %lu %lu", &addr, &num, &l1d_num) != 3)
die("Malformed profile file\n");
summary += num;
l1d_summary += l1d_num;
found = false;
bfd_map_over_sections(bin, findsym, &addr);
@ -161,20 +162,20 @@ int main(int argc, char **argv) {
}
fprintf(out, "fn=%s\n", funcname);
fprintf(out, "%u %lu\n\n", lineno, num);
fprintf(out, "%u %lu %lu\n\n", lineno, num, l1d_num);
} else {
fprintf(out, "fl=??\n");
map<uint32_t, string>::const_iterator it = funcs.lower_bound(addr);
it--;
fprintf(out, "fn=%s\n", it->second.c_str());
fprintf(out, "0 %lu\n\n", num);
fprintf(out, "0 %lu %lu\n\n", num, l1d_num);
}
}
bfd_close(bin);
fprintf(out, "totals: %lu\n", summary);
fprintf(out, "totals: %lu %lu\n", summary, l1d_summary);
fclose(f);
fclose(out);

2
vr4300/cpu.c
View file

@ -66,7 +66,7 @@ int vr4300_init(struct vr4300 *vr4300, struct bus_controller *bus, bool profilin
vr4300->mi_regs[MI_INIT_MODE_REG] = 0x80;
if (profiling)
vr4300->profile_samples = calloc(8 * 1024 * 1024, sizeof(uint64_t));
vr4300->profile_samples = calloc(2 * 8 * 1024 * 1024, sizeof(uint64_t));
else
vr4300->profile_samples = NULL;

13
vr4300/pipeline.c
View file

@ -298,6 +298,12 @@ static int vr4300_dc_stage(struct vr4300 *vr4300) {
request->paddr = paddr;
exdc_latch->cached = cached;
if (vr4300->profile_samples) {
uint32_t idx = exdc_latch->common.pc - 0x80000000;
idx &= (8 * 1024 * 1024) - 1;
vr4300->profile_samples[idx + (8 * 1024 * 1024)]++;
}
// Miss: stall for one cycle, then move to the DCM phase.
vr4300->pipeline.cycles_to_stall = 0;
vr4300->regs[PIPELINE_CYCLE_TYPE] = 6;
@ -316,6 +322,13 @@ static int vr4300_dc_stage(struct vr4300 *vr4300) {
request->paddr = paddr;
exdc_latch->cached = cached;
if (vr4300->profile_samples) {
uint32_t idx = exdc_latch->common.pc - 0x80000000;
idx &= (8 * 1024 * 1024) - 1;
vr4300->profile_samples[idx + (8 * 1024 * 1024)]++;
}
VR4300_DCM(vr4300);
return 1;
}