lkml.org 
[lkml]   [2009]   [Jul]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Subject[PATCH 3/6 -tip] perf_counter: Add Generalized Hardware vectored co-processor support for AMD
From
Date

$ ./perf stat -e add -e multiply -e divide -e vec-idle-cycles -e vec-stall-cycles -e vec-ops -- ls -lR /usr/include/ > /dev/null

Performance counter stats for 'ls -lR /usr/include/':

4218 vec-adds (scaled from 66.60%)
7426 vec-muls (scaled from 66.67%)
5441 vec-divs (scaled from 66.29%)
821982187 vec-idle-cycles (scaled from 66.45%)
2681 vec-stall-cycles (scaled from 67.11%)
7887 vec-ops (scaled from 66.88%)

0.417614573 seconds time elapsed

$ ./perf stat -e add -e multiply -e divide -e vec-idle-cycles -e vec-stall-cycles -e vec-ops -- /usr/bin/rhythmbox ~jaswinder/Music/singhiskinng.mp3

Performance counter stats for '/usr/bin/rhythmbox /home/jaswinder/Music/singhiskinng.mp3':

17552264 vec-adds (scaled from 66.28%)
19715258 vec-muls (scaled from 66.63%)
15862733 vec-divs (scaled from 66.82%)
23735187095 vec-idle-cycles (scaled from 66.89%)
11353159 vec-stall-cycles (scaled from 66.90%)
36628571 vec-ops (scaled from 66.48%)

298.350012843 seconds time elapsed

$ ./perf stat -e add -e multiply -e divide -e vec-idle-cycles -e vec-stall-cycles -e vec-ops -- /usr/bin/vlc ~jaswinder/Videos/Linus_Torvalds_interview_with_Charlie_Rose_Part_1.flv

Performance counter stats for '/usr/bin/vlc /home/jaswinder/Videos/Linus_Torvalds_interview_with_Charlie_Rose_Part_1.flv':

20177177044 vec-adds (scaled from 66.63%)
34101687027 vec-muls (scaled from 66.64%)
3984060862 vec-divs (scaled from 66.71%)
26349684710 vec-idle-cycles (scaled from 66.65%)
9052001905 vec-stall-cycles (scaled from 66.66%)
76440734242 vec-ops (scaled from 66.71%)

272.523058097 seconds time elapsed

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
---
arch/x86/kernel/cpu/perf_counter.c | 33 +++++++++++++++++++++++++++++++
include/linux/perf_counter.h | 15 ++++++++++++++
kernel/perf_counter.c | 1 +
tools/perf/util/parse-events.c | 38 ++++++++++++++++++++++++++++++++++++
4 files changed, 87 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c
index d4cf4ce..8092200 100644
--- a/arch/x86/kernel/cpu/perf_counter.c
+++ b/arch/x86/kernel/cpu/perf_counter.c
@@ -372,6 +372,12 @@ static const u64 atom_hw_cache_event_ids
},
};

+/*
+ * Generalized hw vectored co-processor event table
+ */
+
+static u64 __read_mostly hw_vector_event_ids[PERF_COUNT_HW_VECTOR_MAX];
+
static u64 intel_pmu_raw_event(u64 event)
{
#define CORE_EVNTSEL_EVENT_MASK 0x000000FFULL
@@ -481,6 +487,17 @@ static const u64 amd_hw_cache_event_ids
},
};

+static const u64 amd_hw_vector_event_ids[] =
+{
+ [PERF_COUNT_HW_VECTOR_ADD] = 0x0100, /* Dispatched FPU Add */
+ [PERF_COUNT_HW_VECTOR_MULTIPLY] = 0x0200, /* Dispatched FPU Multiply */
+ [PERF_COUNT_HW_VECTOR_DIVIDE] = 0x0400, /* Dispatched FPU Store */
+ [PERF_COUNT_HW_VECTOR_IDLE_CYCLES] = 0x0001, /* FPU Empty cycles */
+ [PERF_COUNT_HW_VECTOR_STALL_CYCLES] = 0x00D7, /* Dispatch stall for FPU */
+ [PERF_COUNT_HW_VECTOR_OPS] = 0x0FCB, /* Retired x87|(MMX & 3Dnow)
+ |SSE & SSE2) Instructions */
+};
+
/*
* AMD Performance Monitor K7 and later.
*/
@@ -659,6 +676,17 @@ set_ext_hw_attr(struct hw_perf_counter *hwc, struct perf_counter_attr *attr)
return 0;
}

+static inline int
+set_hw_vector_attr(struct hw_perf_counter *hwc, struct perf_counter_attr *attr)
+{
+ if (attr->config >= PERF_COUNT_HW_VECTOR_MAX)
+ return -EINVAL;
+
+ hwc->config |= hw_vector_event_ids[attr->config];
+
+ return 0;
+}
+
/*
* Setup the hardware configuration for a given attr_type
*/
@@ -716,6 +744,9 @@ static int __hw_perf_counter_init(struct perf_counter *counter)
if (attr->type == PERF_TYPE_HW_CACHE)
return set_ext_hw_attr(hwc, attr);

+ if (attr->type == PERF_TYPE_HW_VECTOR)
+ return set_hw_vector_attr(hwc, attr);
+
if (attr->config >= x86_pmu.max_events)
return -EINVAL;
/*
@@ -1468,6 +1499,8 @@ static int amd_pmu_init(void)
/* Events are common for all AMDs */
memcpy(hw_cache_event_ids, amd_hw_cache_event_ids,
sizeof(hw_cache_event_ids));
+ memcpy(hw_vector_event_ids, amd_hw_vector_event_ids,
+ sizeof(hw_vector_event_ids));

return 0;
}
diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index 5e970c7..e91b712 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -31,6 +31,7 @@ enum perf_type_id {
PERF_TYPE_TRACEPOINT = 2,
PERF_TYPE_HW_CACHE = 3,
PERF_TYPE_RAW = 4,
+ PERF_TYPE_HW_VECTOR = 5,

PERF_TYPE_MAX, /* non-ABI */
};
@@ -89,6 +90,20 @@ enum perf_hw_cache_op_result_id {
};

/*
+ * Generalized hardware vectored co-processor counters:
+ */
+enum perf_hw_vector_id {
+ PERF_COUNT_HW_VECTOR_ADD = 0,
+ PERF_COUNT_HW_VECTOR_MULTIPLY = 1,
+ PERF_COUNT_HW_VECTOR_DIVIDE = 2,
+ PERF_COUNT_HW_VECTOR_IDLE_CYCLES = 3,
+ PERF_COUNT_HW_VECTOR_STALL_CYCLES = 4,
+ PERF_COUNT_HW_VECTOR_OPS = 5,
+
+ PERF_COUNT_HW_VECTOR_MAX, /* non-ABI */
+};
+
+/*
* Special "software" counters provided by the kernel, even if the hardware
* does not support performance counters. These counters measure various
* physical and sw events of the kernel (and allow the profiling of them as
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index d55a50d..dd3848a 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -3838,6 +3838,7 @@ perf_counter_alloc(struct perf_counter_attr *attr,
case PERF_TYPE_RAW:
case PERF_TYPE_HARDWARE:
case PERF_TYPE_HW_CACHE:
+ case PERF_TYPE_HW_VECTOR:
pmu = hw_perf_counter_init(counter);
break;

diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 4d042f1..5e5d17e 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -40,6 +40,17 @@ static struct event_symbol event_symbols[] = {
{ CSW(CPU_MIGRATIONS), "cpu-migrations", "migrations" },
};

+#define CHVECTOR(x) .type = PERF_TYPE_HW_VECTOR, .config = PERF_COUNT_HW_VECTOR_##x
+
+static struct event_symbol vector_event_symbols[] = {
+ { CHVECTOR(ADD), "vec-adds", "add" },
+ { CHVECTOR(MULTIPLY), "vec-muls", "multiply" },
+ { CHVECTOR(DIVIDE), "vec-divs", "divide" },
+ { CHVECTOR(IDLE_CYCLES), "vec-idle-cycles", "vec-empty-cycles"},
+ { CHVECTOR(STALL_CYCLES), "vec-stall-cycles", "vec-busy-cycles"},
+ { CHVECTOR(OPS), "vec-ops", "vec-operations"},
+};
+
#define __PERF_COUNTER_FIELD(config, name) \
((config & PERF_COUNTER_##name##_MASK) >> PERF_COUNTER_##name##_SHIFT)

@@ -172,6 +183,11 @@ char *event_name(int counter)
return event_cache_name(cache_type, cache_op, cache_result);
}

+ case PERF_TYPE_HW_VECTOR:
+ if (config < PERF_COUNT_HW_VECTOR_MAX)
+ return vector_event_symbols[config].symbol;
+ return "unknown-vector";
+
case PERF_TYPE_SOFTWARE:
if (config < PERF_COUNT_SW_MAX)
return sw_event_names[config];
@@ -250,6 +266,19 @@ static int check_events(const char *str, unsigned int i)
return 0;
}

+static int check_vector_events(const char *str, unsigned int i)
+{
+ if (!strncmp(str, vector_event_symbols[i].symbol,
+ strlen(vector_event_symbols[i].symbol)))
+ return 1;
+
+ if (strlen(vector_event_symbols[i].alias))
+ if (!strncmp(str, vector_event_symbols[i].alias,
+ strlen(vector_event_symbols[i].alias)))
+ return 1;
+ return 0;
+}
+
/*
* Each event can have multiple symbolic names.
* Symbolic names are (almost) exactly matched.
@@ -297,6 +326,15 @@ static int parse_event_symbols(const char *str, struct perf_counter_attr *attr)
}
}

+ for (i = 0; i < ARRAY_SIZE(vector_event_symbols); i++) {
+ if (check_vector_events(str, i)) {
+ attr->type = vector_event_symbols[i].type;
+ attr->config = vector_event_symbols[i].config;
+
+ return 0;
+ }
+ }
+
return parse_generic_hw_symbols(str, attr);
}

--
1.6.0.6




\
 
 \ /
  Last update: 2009-07-01 11:51    [W:0.088 / U:0.568 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site