[Intel-gfx] [PATCH i-g-t 06/12] tools/intel_gpu_top: Optimise interactive display a bit

[Intel-gfx] [PATCH i-g-t 06/12] tools/intel_gpu_top: Optimise interactive display a bit

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Padding the percentage bars and table columns with spaces happens quite a
lot so lets do better than putchar at a time. Have a table of visually
empty strings and build the required length out of those chunks.

While at it, also move the percentage bar table into its function scope.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/intel_gpu_top.c | 38 +++++++++++++++++++++++++++++++++-----
 1 file changed, 33 insertions(+), 5 deletions(-)

diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 6363f460c892..a7a24e527f01 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -928,14 +928,40 @@ static void free_display_clients(struct igt_drm_clients *clients)
 	free(clients);
 }
 
-static const char *bars[] = { " ", "▏", "▎", "▍", "▌", "▋", "▊", "▉", "█" };
-
 static unsigned int n_spaces(const unsigned int n)
 {
-	unsigned int i;
+	static const char *spaces[] = {
+		" ",
+		"  ",
+		"   ",
+		"    ",
+		"     ",
+		"      ",
+		"       ",
+		"        ",
+		"         ",
+		"          ",
+		"           ",
+		"            ",
+		"             ",
+		"              ",
+		"               ",
+		"                ",
+		"                 ",
+		"                  ",
+		"                   ",
+#define MAX_SPACES 19
+	};
+	unsigned int i, r = n;
 
-	for (i = 0; i < n; i++)
-		putchar(' ');
+	while (r) {
+		if (r > MAX_SPACES)
+			i = MAX_SPACES - 1;
+		else
+			i = r - 1;
+		fputs(spaces[i], stdout);
+		r -= i + 1;
+	}
 
 	return n;
 }
@@ -943,6 +969,8 @@ static unsigned int n_spaces(const unsigned int n)
 static void
 print_percentage_bar(double percent, double max, int max_len, bool numeric)
 {
+	static const char *bars[] =
+		{ " ", "▏", "▎", "▍", "▌", "▋", "▊", "▉", "█" };
 	int bar_len, i, len = max_len - 2;
 	const int w = 8;
 
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 00/12] fdinfo tests, intel_gpu_top memory support, etc

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Some basic testst for fdinfo memory stats, intel_gpu_top memory stats support
(first draft) and a couple fixlets.

Tvrtko Ursulin (12):
  tests/i915/drm_fdinfo: Check engine info is supported
  tests/i915/drm_fdinfo: Stress test context close versus fdinfo reads
  tests/i915/drm_fdinfo: Add some memory info tests
  tools/intel_gpu_top: Fix clients header width when no clients
  tools/intel_gpu_top: Fix client layout on first sample period
  tools/intel_gpu_top: Optimise interactive display a bit
  lib/igt_drm_fdinfo: Copy over region map name on match
  lib/igt_drm_clients: Fix client id type confusion
  lib/igt_drm_clients: Allow passing in the memory region map
  tools/intel_gpu_top: Fully wrap clients operations
  tools/intel_gpu_top: Add per client memory info
  tools/intel_gpu_top: Add ability to show memory region breakdown

 lib/igt_drm_clients.c    |   7 +-
 lib/igt_drm_clients.h    |   5 +-
 lib/igt_drm_fdinfo.c     |   4 +
 man/intel_gpu_top.rst    |   4 +
 tests/intel/drm_fdinfo.c | 286 +++++++++++++++++++++++++++++++++++++++
 tools/gputop.c           |   4 +-
 tools/intel_gpu_top.c    | 250 +++++++++++++++++++++++++++++-----
 7 files changed, 518 insertions(+), 42 deletions(-)

-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 01/12] tests/i915/drm_fdinfo: Check engine info is supported

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

On top of checking that parsing works, check that there are some engines
present. This will be needed once the memory stats are added and so return
value from __igt_parse_drm_fdinfo() will then be possible to be greater
than zero even when engine stats are not supported.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tests/intel/drm_fdinfo.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tests/intel/drm_fdinfo.c b/tests/intel/drm_fdinfo.c
index aca19db50680..344c44dce78b 100644
--- a/tests/intel/drm_fdinfo.c
+++ b/tests/intel/drm_fdinfo.c
@@ -737,6 +737,7 @@ igt_main
 
 		igt_require_gem(i915);
 		igt_require(igt_parse_drm_fdinfo(i915, &info, NULL, 0, NULL, 0));
+		igt_require(info.num_engines);
 
 		ctx = intel_ctx_create_all_physical(i915);
 
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 02/12] tests/i915/drm_fdinfo: Stress test context close versus fdinfo reads

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

A short smoke tests to exercise fdinfo reads in parallel to contexts
getting created and destroyed.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tests/intel/drm_fdinfo.c | 68 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 68 insertions(+)

diff --git a/tests/intel/drm_fdinfo.c b/tests/intel/drm_fdinfo.c
index 344c44dce78b..c4218b0d16e6 100644
--- a/tests/intel/drm_fdinfo.c
+++ b/tests/intel/drm_fdinfo.c
@@ -22,11 +22,14 @@
  *
  */
 
+#include <fcntl.h>
+
 #include "igt.h"
 #include "igt_core.h"
 #include "igt_device.h"
 #include "igt_drm_fdinfo.h"
 #include "i915/gem.h"
+#include "i915/gem_create.h"
 #include "i915/gem_vm.h"
 #include "intel_ctx.h"
 /**
@@ -72,6 +75,8 @@
  * SUBTEST: virtual-busy-idle-all
  *
  * SUBTEST: virtual-idle
+ *
+ * SUBTEST: context-close-stress
  */
 
 IGT_TEST_DESCRIPTION("Test the i915 drm fdinfo data");
@@ -717,6 +722,56 @@ virtual_all(int i915, const intel_ctx_cfg_t *base_cfg, unsigned int flags)
 	}
 }
 
+static void stress_context_close(int i915)
+{
+	const uint32_t bbe = MI_BATCH_BUFFER_END;
+	struct igt_helper_process reader = { };
+	struct drm_client_fdinfo info;
+	uint32_t batch;
+	int dir, ret;
+	char buf[64];
+
+	ret = snprintf(buf, sizeof(buf), "%u", i915);
+	igt_assert(ret > 0 && ret < sizeof(buf));
+
+	dir = open("/proc/self/fdinfo", O_DIRECTORY | O_RDONLY);
+	igt_assert_fd(dir);
+
+	memset(&info, 0, sizeof(info));
+	ret = __igt_parse_drm_fdinfo(dir, buf, &info, NULL, 0, NULL, 0);
+	igt_assert(ret > 0);
+	igt_require(info.num_regions);
+
+	batch = gem_create(i915, 4096);
+	gem_write(i915, batch, 0, &bbe, sizeof(bbe));
+
+	igt_fork_helper(&reader) {
+		for (;;) {
+			memset(&info, 0, sizeof(info));
+			ret = __igt_parse_drm_fdinfo(dir, buf, &info,
+						     NULL, 0, NULL, 0);
+			igt_assert(ret > 0);
+		}
+	}
+
+	igt_until_timeout(10) {
+		struct drm_i915_gem_exec_object2 obj = {
+			.handle = batch,
+		};
+		struct drm_i915_gem_execbuffer2 eb = {
+			.buffers_ptr = to_user_pointer(&obj),
+			.buffer_count = 1,
+		};
+
+		eb.rsvd1 = gem_context_create(i915);
+		igt_assert(eb.rsvd1);
+		gem_execbuf(i915, &eb);
+		gem_context_destroy(i915, eb.rsvd1);
+	}
+
+	igt_stop_helper(&reader);
+}
+
 #define test_each_engine(T, i915, ctx, e) \
 	igt_subtest_with_dynamic(T) for_each_ctx_engine(i915, ctx, e) \
 		igt_dynamic_f("%s", e->name)
@@ -848,6 +903,19 @@ igt_main
 	test_each_engine("isolation", i915, ctx, e)
 		single(i915, ctx, e, TEST_BUSY | TEST_ISOLATION);
 
+	igt_subtest_group {
+		int newfd;
+
+		igt_fixture
+			newfd = drm_reopen_driver(i915);
+
+		igt_subtest("context-close-stress")
+			stress_context_close(newfd);
+
+		igt_fixture
+			drm_close_driver(newfd);
+	}
+
 	igt_fixture {
 		intel_ctx_destroy(i915, ctx);
 		drm_close_driver(i915);
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 03/12] tests/i915/drm_fdinfo: Add some memory info tests

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

A few basic smoke tests to check per client memory info looks legit.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tests/intel/drm_fdinfo.c | 217 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 217 insertions(+)

diff --git a/tests/intel/drm_fdinfo.c b/tests/intel/drm_fdinfo.c
index c4218b0d16e6..3ced658f2faa 100644
--- a/tests/intel/drm_fdinfo.c
+++ b/tests/intel/drm_fdinfo.c
@@ -23,6 +23,7 @@
  */
 
 #include <fcntl.h>
+#include <sys/ioctl.h>
 
 #include "igt.h"
 #include "igt_core.h"
@@ -76,6 +77,16 @@
  *
  * SUBTEST: virtual-idle
  *
+ * SUBTEST: memory-info-idle
+ *
+ * SUBTEST: memory-info-active
+ *
+ * SUBTEST: memory-info-resident
+ *
+ * SUBTEST: memory-info-purgeable
+ *
+ * SUBTEST: memory-info-shared
+ *
  * SUBTEST: context-close-stress
  */
 
@@ -143,6 +154,11 @@ static unsigned int measured_usleep(unsigned int usec)
 #define FLAG_HANG (8)
 #define TEST_ISOLATION (16)
 
+#define TEST_ACTIVE TEST_BUSY
+#define TEST_RESIDENT (32)
+#define TEST_PURGEABLE (64)
+#define TEST_SHARED (128)
+
 static void end_spin(int fd, igt_spin_t *spin, unsigned int flags)
 {
 	if (!spin)
@@ -772,6 +788,172 @@ static void stress_context_close(int i915)
 	igt_stop_helper(&reader);
 }
 
+static size_t read_fdinfo(char *buf, const size_t sz, int at, const char *name)
+{
+	size_t count;
+	int fd;
+
+	fd = openat(at, name, O_RDONLY);
+	if (fd < 0)
+		return 0;
+
+	count = read(fd, buf, sz - 1);
+	if (count > 0)
+		buf[count - 1] = 0;
+	close(fd);
+
+	return count > 0 ? count : 0;
+}
+
+/*
+ * At least this much, but maybe less if we started with a driver internal
+ * baseline which can go away behind our back.
+ */
+#define fdinfo_assert_gte(cur, prev, sz, base) \
+({ \
+	int64_t __sz = (sz) - (base); \
+	int64_t __d = (cur) - (prev); \
+	igt_assert_f(__d >= __sz, \
+		     "prev=%"PRIu64" cur=%"PRIu64" delta=%"PRId64" sz=%"PRIu64" baseline=%"PRIu64"\n%s\n", \
+		     (prev), (cur), __d, (sz), (base), fdinfo_buf); \
+})
+
+#define fdinfo_assert_eq(cur, prev, sz, base) \
+({ \
+	int64_t __d = (cur) - (prev); \
+	igt_assert_f(__d == 0, \
+		     "prev=%"PRIu64" cur=%"PRIu64" delta=%"PRId64" sz=%"PRIu64" baseline=%"PRIu64"\n%s\n", \
+		     (prev), (cur), __d, (sz), (base), fdinfo_buf); \
+})
+
+static void
+test_memory(int i915, struct gem_memory_region *mr, unsigned int flags)
+{
+	const unsigned int r = mr->ci.memory_class == I915_MEMORY_CLASS_SYSTEM ? 0 : 1; /* See region map */
+	const uint64_t max_mem = 512ull * 1024 * 1024;
+	const uint64_t max_bo = 16ull * 1024 * 1024;
+	struct drm_client_fdinfo base_info, prev_info = { };
+	struct drm_client_fdinfo info = { };
+	char buf[64], fdinfo_buf[4096];
+	igt_spin_t *spin = NULL;
+	uint64_t total = 0, sz;
+	uint64_t ahnd;
+	int ret, dir;
+
+	i915 = drm_reopen_driver(i915);
+
+	ahnd = get_reloc_ahnd(i915, 0);
+
+	ret = snprintf(buf, sizeof(buf), "%u", i915);
+	igt_assert(ret > 0 && ret < sizeof(buf));
+
+	dir = open("/proc/self/fdinfo", O_DIRECTORY | O_RDONLY);
+	igt_assert_fd(dir);
+
+	gem_quiescent_gpu(i915);
+	ret =  __igt_parse_drm_fdinfo(dir, buf, &info, NULL, 0, NULL, 0);
+	igt_assert(ret > 0);
+	igt_require(info.num_regions);
+	memcpy(&prev_info, &info, sizeof(info));
+	memcpy(&base_info, &info, sizeof(info));
+
+	while (total < max_mem) {
+		static const char *region_map[] = {
+			"system0",
+			"local0",
+		};
+		uint32_t bo;
+
+		sz = random() % max_bo;
+		ret = __gem_create_in_memory_region_list(i915, &bo, &sz, 0,
+							 &mr->ci, 1);
+		igt_assert_eq(ret, 0);
+		total += sz;
+
+		if (flags & (TEST_RESIDENT | TEST_PURGEABLE | TEST_ACTIVE))
+			spin = igt_spin_new(i915,
+					    .dependency = bo,
+					    .ahnd = ahnd);
+		else
+			spin = NULL;
+
+		if (flags & TEST_PURGEABLE) {
+			gem_madvise(i915, bo, I915_MADV_DONTNEED);
+			igt_spin_free(i915, spin);
+			gem_quiescent_gpu(i915);
+			spin = NULL;
+		}
+
+		if (flags & TEST_SHARED) {
+			struct drm_gem_open open_struct;
+			struct drm_gem_flink flink;
+
+			flink.handle = bo;
+			ret = ioctl(i915, DRM_IOCTL_GEM_FLINK, &flink);
+			igt_assert_eq(ret, 0);
+
+			open_struct.name = flink.name;
+			ret = ioctl(i915, DRM_IOCTL_GEM_OPEN, &open_struct);
+			igt_assert_eq(ret, 0);
+			igt_assert(open_struct.handle != 0);
+		}
+
+		memset(&info, 0, sizeof(info));
+		ret =  __igt_parse_drm_fdinfo(dir, buf, &info,
+					      NULL, 0,
+					      region_map, ARRAY_SIZE(region_map));
+		igt_assert(ret > 0);
+		igt_assert(info.num_regions);
+
+		read_fdinfo(fdinfo_buf, sizeof(fdinfo_buf), dir, buf);
+
+		/* >= to account for objects out of our control */
+		fdinfo_assert_gte(info.region_mem[r].total,
+				  prev_info.region_mem[r].total,
+				  sz,
+				  base_info.region_mem[r].total);
+
+		if (flags & TEST_SHARED)
+			fdinfo_assert_gte(info.region_mem[r].shared,
+					  prev_info.region_mem[r].shared,
+					  sz,
+					  base_info.region_mem[r].shared);
+		else
+			fdinfo_assert_eq(info.region_mem[r].shared,
+					 prev_info.region_mem[r].shared,
+					 sz,
+					 base_info.region_mem[r].shared);
+
+		if (flags & (TEST_RESIDENT | TEST_PURGEABLE | TEST_ACTIVE))
+			fdinfo_assert_gte(info.region_mem[r].resident,
+					  (uint64_t)0, /* We can only be sure the current buffer is resident. */
+					  sz,
+					  (uint64_t)0);
+
+		if (flags & TEST_PURGEABLE)
+			fdinfo_assert_gte(info.region_mem[r].purgeable,
+					  (uint64_t)0, /* We can only be sure the current buffer is purgeable (subset of resident). */
+					  sz,
+					  (uint64_t)0);
+
+		if (flags & TEST_ACTIVE)
+			fdinfo_assert_gte(info.region_mem[r].active,
+					  (uint64_t)0, /* We can only be sure the current buffer is active. */
+					  sz,
+					  (uint64_t)0);
+
+		memcpy(&prev_info, &info, sizeof(info));
+
+		if (spin) {
+			igt_spin_free(i915, spin);
+			gem_quiescent_gpu(i915);
+		}
+	}
+
+	put_ahnd(ahnd);
+	close(i915);
+}
+
 #define test_each_engine(T, i915, ctx, e) \
 	igt_subtest_with_dynamic(T) for_each_ctx_engine(i915, ctx, e) \
 		igt_dynamic_f("%s", e->name)
@@ -903,6 +1085,41 @@ igt_main
 	test_each_engine("isolation", i915, ctx, e)
 		single(i915, ctx, e, TEST_BUSY | TEST_ISOLATION);
 
+	igt_subtest_with_dynamic("memory-info-idle") {
+		for_each_memory_region(r, i915) {
+			igt_dynamic_f("%s", r->name)
+				test_memory(i915, r, 0);
+		}
+	}
+
+	igt_subtest_with_dynamic("memory-info-resident") {
+		for_each_memory_region(r, i915) {
+			igt_dynamic_f("%s", r->name)
+				test_memory(i915, r, TEST_RESIDENT);
+		}
+	}
+
+	igt_subtest_with_dynamic("memory-info-purgeable") {
+		for_each_memory_region(r, i915) {
+			igt_dynamic_f("%s", r->name)
+				test_memory(i915, r, TEST_PURGEABLE);
+		}
+	}
+
+	igt_subtest_with_dynamic("memory-info-active") {
+		for_each_memory_region(r, i915) {
+			igt_dynamic_f("%s", r->name)
+				test_memory(i915, r, TEST_ACTIVE);
+		}
+	}
+
+	igt_subtest_with_dynamic("memory-info-shared") {
+		for_each_memory_region(r, i915) {
+			igt_dynamic_f("%s", r->name)
+				test_memory(i915, r, TEST_SHARED);
+		}
+	}
+
 	igt_subtest_group {
 		int newfd;
 
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 04/12] tools/intel_gpu_top: Fix clients header width when no clients

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Recent refactoring broke the clients header in cases when there are no
clients displayed. To fix it we need to account the width of the "NAME"
label.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/intel_gpu_top.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 87e9681e53b4..66b76c4afed0 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -1965,6 +1965,8 @@ print_clients_header(struct igt_drm_clients *clients, int lines,
 		     int con_w, int con_h, int *class_w)
 {
 	struct intel_clients *iclients = clients->private_data;
+	const int max_name_len = clients->max_name_len < 4 ?
+				 4 : clients->max_name_len; /* At least "NAME" */
 
 	if (output_mode == INTERACTIVE) {
 		unsigned int num_active = 0;
@@ -1988,9 +1990,8 @@ print_clients_header(struct igt_drm_clients *clients, int lines,
 					num_active++;
 			}
 
-			*class_w = width =
-				(con_w - len - clients->max_name_len - 1) /
-				num_active;
+			*class_w = width = (con_w - len - max_name_len - 1) /
+					   num_active;
 
 			for (i = 0; i <= iclients->classes.max_engine_id; i++) {
 				const char *name = iclients->classes.names[i];
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 05/12] tools/intel_gpu_top: Fix client layout on first sample period

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

When I moved the client name to be last, I did not account for the fact
current code skips showing engine utilisation until at least two sampling
periods have passed. Consequence of this is that client name gets printed
as the second field and not under the "NAME" column header.

Fix it by emitting spaces instead of engine utilisation until two samples
have been collected.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/intel_gpu_top.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 66b76c4afed0..de37ff603b6f 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -926,12 +926,14 @@ static void free_display_clients(struct igt_drm_clients *clients)
 
 static const char *bars[] = { " ", "▏", "▎", "▍", "▌", "▋", "▊", "▉", "█" };
 
-static void n_spaces(const unsigned int n)
+static unsigned int n_spaces(const unsigned int n)
 {
 	unsigned int i;
 
 	for (i = 0; i < n; i++)
 		putchar(' ');
+
+	return n;
 }
 
 static void
@@ -2041,14 +2043,17 @@ print_client(struct igt_drm_client *c, struct engines *engines, double t, int li
 
 		len = printf("%*s ", clients->max_pid_len, c->pid_str);
 
-		for (i = 0;
-		     c->samples > 1 && i <= iclients->classes.max_engine_id;
-		     i++) {
+		for (i = 0; i <= iclients->classes.max_engine_id; i++) {
 			double pct, max;
 
 			if (!iclients->classes.capacity[i])
 				continue;
 
+			if (c->samples < 2) {
+				len += n_spaces(*class_w);
+				continue;
+			}
+
 			pct = (double)c->val[i] / period_us / 1e3 * 100;
 
 			/*
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 06/12] tools/intel_gpu_top: Optimise interactive display a bit

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Padding the percentage bars and table columns with spaces happens quite a
lot so lets do better than putchar at a time. Have a table of visually
empty strings and build the required length out of those chunks.

While at it, also move the percentage bar table into its function scope.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/intel_gpu_top.c | 38 +++++++++++++++++++++++++++++++++-----
 1 file changed, 33 insertions(+), 5 deletions(-)

diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index de37ff603b6f..739ed9839961 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -924,14 +924,40 @@ static void free_display_clients(struct igt_drm_clients *clients)
 	free(clients);
 }
 
-static const char *bars[] = { " ", "▏", "▎", "▍", "▌", "▋", "▊", "▉", "█" };
-
 static unsigned int n_spaces(const unsigned int n)
 {
-	unsigned int i;
+	static const char *spaces[] = {
+		" ",
+		"  ",
+		"   ",
+		"    ",
+		"     ",
+		"      ",
+		"       ",
+		"        ",
+		"         ",
+		"          ",
+		"           ",
+		"            ",
+		"             ",
+		"              ",
+		"               ",
+		"                ",
+		"                 ",
+		"                  ",
+		"                   ",
+#define MAX_SPACES 19
+	};
+	unsigned int i, r = n;
 
-	for (i = 0; i < n; i++)
-		putchar(' ');
+	while (r) {
+		if (r > MAX_SPACES)
+			i = MAX_SPACES - 1;
+		else
+			i = r - 1;
+		fputs(spaces[i], stdout);
+		r -= i + 1;
+	}
 
 	return n;
 }
@@ -939,6 +965,8 @@ static unsigned int n_spaces(const unsigned int n)
 static void
 print_percentage_bar(double percent, double max, int max_len, bool numeric)
 {
+	static const char *bars[] =
+		{ " ", "▏", "▎", "▍", "▌", "▋", "▊", "▉", "█" };
 	int bar_len, i, len = max_len - 2;
 	const int w = 8;
 
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 07/12] lib/igt_drm_fdinfo: Copy over region map name on match

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

I will need some record of which regions were found for intel_gpu_top so
lets just copy over the region name from the map on the first match.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 lib/igt_drm_fdinfo.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/lib/igt_drm_fdinfo.c b/lib/igt_drm_fdinfo.c
index b72822894782..222ccbfb1fd6 100644
--- a/lib/igt_drm_fdinfo.c
+++ b/lib/igt_drm_fdinfo.c
@@ -148,6 +148,10 @@ static int parse_region(char *line, struct drm_client_fdinfo *info,
 		for (i = 0; i < region_entries; i++) {
 			if (!strncmp(name, region_map[i], name_len)) {
 				found = i;
+				if (!info->region_names[info->num_regions][0]) {
+					assert(name_len < sizeof(info->region_names[i]));
+					strncpy(info->region_names[i], name, name_len);
+				}
 				break;
 			}
 		}
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 08/12] lib/igt_drm_clients: Fix client id type confusion

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Igt_drm_fdinfo defines it as an unsigned long so it is best that it
matches here as well.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 lib/igt_drm_clients.c | 2 +-
 lib/igt_drm_clients.h | 2 +-
 tools/intel_gpu_top.c | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/lib/igt_drm_clients.c b/lib/igt_drm_clients.c
index 47ad137d5f1f..da51d7335b2b 100644
--- a/lib/igt_drm_clients.c
+++ b/lib/igt_drm_clients.c
@@ -49,7 +49,7 @@ struct igt_drm_clients *igt_drm_clients_init(void *private_data)
 static struct igt_drm_client *
 igt_drm_clients_find(struct igt_drm_clients *clients,
 		     enum igt_drm_client_status status,
-		     unsigned int drm_minor, unsigned int id)
+		     unsigned int drm_minor, unsigned long id)
 {
 	unsigned int start, num;
 	struct igt_drm_client *c;
diff --git a/lib/igt_drm_clients.h b/lib/igt_drm_clients.h
index 07bd236d43bf..cd37f8508b20 100644
--- a/lib/igt_drm_clients.h
+++ b/lib/igt_drm_clients.h
@@ -56,7 +56,7 @@ struct igt_drm_client {
 	enum igt_drm_client_status status;
 	struct igt_drm_client_regions *regions; /* Memory regions present in this client, to map with memory usage. */
 	struct igt_drm_client_engines *engines; /* Engines used by this client, to map with busynees data. */
-	unsigned int id; /* DRM client id from fdinfo. */
+	unsigned long id; /* DRM client id from fdinfo. */
 	unsigned int drm_minor; /* DRM minor of this client. */
 	unsigned int pid; /* PID which has this DRM fd open. */
 	char pid_str[10]; /* Cached PID representation. */
diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 739ed9839961..35c125f22c48 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -2102,7 +2102,7 @@ print_client(struct igt_drm_client *c, struct engines *engines, double t, int li
 	} else if (output_mode == JSON) {
 		char buf[64];
 
-		snprintf(buf, sizeof(buf), "%u", c->id);
+		snprintf(buf, sizeof(buf), "%lu", c->id);
 		pops->open_struct(buf);
 
 		__json_add_member("name", c->print_name);
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 09/12] lib/igt_drm_clients: Allow passing in the memory region map

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Same concept as with the engine map, allowing callers to pass in fixed
map of names to indices, simplifying their implementation and avoiding
auto-detection while parsing.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 lib/igt_drm_clients.c | 5 +++--
 lib/igt_drm_clients.h | 3 ++-
 tools/gputop.c        | 4 ++--
 tools/intel_gpu_top.c | 8 +++++---
 4 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/lib/igt_drm_clients.c b/lib/igt_drm_clients.c
index da51d7335b2b..025d60c51503 100644
--- a/lib/igt_drm_clients.c
+++ b/lib/igt_drm_clients.c
@@ -445,7 +445,8 @@ struct igt_drm_clients *
 igt_drm_clients_scan(struct igt_drm_clients *clients,
 		     bool (*filter_client)(const struct igt_drm_clients *,
 					   const struct drm_client_fdinfo *),
-		     const char **name_map, unsigned int map_entries)
+		     const char **name_map, unsigned int map_entries,
+		     const char **region_map, unsigned int region_entries)
 {
 	struct dirent *proc_dent;
 	struct igt_drm_client *c;
@@ -524,7 +525,7 @@ igt_drm_clients_scan(struct igt_drm_clients *clients,
 			if (!__igt_parse_drm_fdinfo(dirfd(fdinfo_dir),
 						    fdinfo_dent->d_name, &info,
 						    name_map, map_entries,
-						    NULL, 0))
+						    region_map, region_entries))
 				continue;
 
 			if (filter_client && !filter_client(clients, &info))
diff --git a/lib/igt_drm_clients.h b/lib/igt_drm_clients.h
index cd37f8508b20..52888aedc25a 100644
--- a/lib/igt_drm_clients.h
+++ b/lib/igt_drm_clients.h
@@ -93,7 +93,8 @@ struct igt_drm_clients *
 igt_drm_clients_scan(struct igt_drm_clients *clients,
 		     bool (*filter_client)(const struct igt_drm_clients *,
 					   const struct drm_client_fdinfo *),
-		     const char **name_map, unsigned int map_entries);
+		     const char **name_map, unsigned int map_entries,
+		     const char **region_map, unsigned int region_entries);
 
 struct igt_drm_clients *
 igt_drm_clients_sort(struct igt_drm_clients *clients,
diff --git a/tools/gputop.c b/tools/gputop.c
index ea95e0333dd2..71e28f43ee4c 100644
--- a/tools/gputop.c
+++ b/tools/gputop.c
@@ -253,7 +253,7 @@ int main(int argc, char **argv)
 	if (!clients)
 		exit(1);
 
-	igt_drm_clients_scan(clients, NULL, NULL, 0);
+	igt_drm_clients_scan(clients, NULL, NULL, 0, NULL, 0);
 
 	for (;;) {
 		struct igt_drm_client *c, *prevc = NULL;
@@ -270,7 +270,7 @@ int main(int argc, char **argv)
 			}
 		}
 
-		igt_drm_clients_scan(clients, NULL, NULL, 0);
+		igt_drm_clients_scan(clients, NULL, NULL, 0, NULL, 0);
 		igt_drm_clients_sort(clients, client_cmp);
 
 		printf("\033[H\033[J");
diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 35c125f22c48..5d44ba45335e 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -2615,8 +2615,9 @@ int main(int argc, char **argv)
 	}
 
 	pmu_sample(engines);
-	igt_drm_clients_scan(clients, client_match, engine_map,
-			     ARRAY_SIZE(engine_map));
+	igt_drm_clients_scan(clients, client_match,
+			     engine_map, ARRAY_SIZE(engine_map),
+			     NULL, 0);
 	gettime(&ts);
 
 	if (output_mode == JSON)
@@ -2651,7 +2652,8 @@ int main(int argc, char **argv)
 			display_clients(igt_drm_clients_scan(clients,
 							     client_match,
 							     engine_map,
-							     ARRAY_SIZE(engine_map)));
+							     ARRAY_SIZE(engine_map),
+							     NULL, 0));
 		scan_us = elapsed_us(&ts, period_us);
 
 		if (stop_top)
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 10/12] tools/intel_gpu_top: Fully wrap clients operations

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Wrap all operations on clients via the Intel specific wrappers in order to
simplify upcoming work.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/intel_gpu_top.c | 42 ++++++++++++++++++++++--------------------
 1 file changed, 22 insertions(+), 20 deletions(-)

diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 5d44ba45335e..4b9ef306ca8e 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -132,6 +132,7 @@ struct engines {
 struct intel_clients {
 	const char *pci_slot;
 	struct igt_drm_client_engines classes;
+	struct igt_drm_clients *clients;
 };
 
 static struct termios termios_orig;
@@ -2432,19 +2433,22 @@ intel_init_clients(struct intel_clients *iclients,
 		iclients->classes.capacity[i] = engines->class[i].num_engines;
 		iclients->classes.names[i] = strdup(engines->class[i].name);
 	}
+
+	iclients->clients = igt_drm_clients_init(iclients);
 }
 
 static void intel_free_clients(struct intel_clients *iclients)
 {
+	if (iclients->clients)
+		igt_drm_clients_free(iclients->clients);
+
 	free((void *)iclients->pci_slot);
 	free(iclients->classes.capacity);
 	free(iclients->classes.names);
 }
 
-int main(int argc, char **argv)
+static void intel_scan_clients(struct intel_clients *iclients)
 {
-	unsigned int period_us = DEFAULT_PERIOD_MS * 1000;
-	struct igt_drm_clients *clients = NULL;
 	static const char *engine_map[] = {
 		"render",
 		"copy",
@@ -2452,6 +2456,15 @@ int main(int argc, char **argv)
 		"video-enhance",
 		"compute",
 	};
+
+	igt_drm_clients_scan(iclients->clients, client_match,
+			     engine_map, ARRAY_SIZE(engine_map),
+			     NULL, 0);
+}
+
+int main(int argc, char **argv)
+{
+	unsigned int period_us = DEFAULT_PERIOD_MS * 1000;
 	bool physical_engines = false;
 	struct intel_clients iclients;
 	int con_w = -1, con_h = -1;
@@ -2609,15 +2622,11 @@ int main(int argc, char **argv)
 
 	init_engine_classes(engines);
 
-	if (has_drm_fdinfo(&card)) {
+	if (has_drm_fdinfo(&card))
 		intel_init_clients(&iclients, &card, engines);
-		clients = igt_drm_clients_init(&iclients);
-	}
 
 	pmu_sample(engines);
-	igt_drm_clients_scan(clients, client_match,
-			     engine_map, ARRAY_SIZE(engine_map),
-			     NULL, 0);
+	intel_scan_clients(&iclients);
 	gettime(&ts);
 
 	if (output_mode == JSON)
@@ -2648,12 +2657,8 @@ int main(int argc, char **argv)
 		pmu_sample(engines);
 		t = (double)(engines->ts.cur - engines->ts.prev) / 1e9;
 
-		disp_clients =
-			display_clients(igt_drm_clients_scan(clients,
-							     client_match,
-							     engine_map,
-							     ARRAY_SIZE(engine_map),
-							     NULL, 0));
+		intel_scan_clients(&iclients);
+		disp_clients = display_clients(iclients.clients);
 		scan_us = elapsed_us(&ts, period_us);
 
 		if (stop_top)
@@ -2704,7 +2709,7 @@ int main(int argc, char **argv)
 			pops->close_struct();
 		}
 
-		if (disp_clients != clients)
+		if (disp_clients != iclients.clients)
 			free_display_clients(disp_clients);
 
 		if (stop_top)
@@ -2719,10 +2724,7 @@ int main(int argc, char **argv)
 	if (output_mode == JSON)
 		printf("]\n");
 
-	if (clients) {
-		igt_drm_clients_free(clients);
-		intel_free_clients(&iclients);
-	}
+	intel_free_clients(&iclients);
 
 	free(codename);
 err_pmu:
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 11/12] tools/intel_gpu_top: Add per client memory info

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

JSON output has the full breakdown but for now the interactive mode only
shows total and resident aggregated for all memory regions.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/intel_gpu_top.c | 114 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 112 insertions(+), 2 deletions(-)

diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 4b9ef306ca8e..958df42e8c20 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -133,11 +133,24 @@ struct intel_clients {
 	const char *pci_slot;
 	struct igt_drm_client_engines classes;
 	struct igt_drm_clients *clients;
+	struct igt_drm_client_regions *regions; /* Borrowed from first client */
 };
 
 static struct termios termios_orig;
 static bool class_view;
 
+/* Maps i915 fdinfo names to indices */
+static const char *memory_region_map[] = {
+	"system0",
+	"local0",
+};
+
+/* For JSON, 1:1 with indices above. */
+static const char *json_memory_region_names[] = {
+	"system",
+	"local",
+};
+
 __attribute__((format(scanf,3,4)))
 static int igt_sysfs_scanf(int dir, const char *attr, const char *fmt, ...)
 {
@@ -880,6 +893,9 @@ static struct igt_drm_clients *display_clients(struct igt_drm_clients *clients)
 			ac->val = calloc(c->engines->max_engine_id + 1,
 					 sizeof(ac->val[0]));
 			assert(ac->val);
+			ac->regions = c->regions;
+			ac->memory = calloc(c->regions->max_region_id + 1,
+					    sizeof(ac->memory[0]));
 			ac->samples = 1;
 		}
 
@@ -894,6 +910,14 @@ static struct igt_drm_clients *display_clients(struct igt_drm_clients *clients)
 
 		for (i = 0; i <= c->engines->max_engine_id; i++)
 			ac->val[i] += c->val[i];
+
+		for (i = 0; i <= c->regions->max_region_id; i++) {
+			ac->memory[i].total += c->memory[i].total;
+			ac->memory[i].shared += c->memory[i].shared;
+			ac->memory[i].resident += c->memory[i].resident;
+			ac->memory[i].purgeable += c->memory[i].purgeable;
+			ac->memory[i].active += c->memory[i].active;
+		}
 	}
 
 	aggregated->num_clients = num;
@@ -918,8 +942,10 @@ static void free_display_clients(struct igt_drm_clients *clients)
 	 * "display" clients are not proper clients and have un-initialized
 	 * or borrowed fields which we don't want the library to try and free.
 	 */
-	igt_for_each_drm_client(clients, c, tmp)
+	igt_for_each_drm_client(clients, c, tmp) {
 		free(c->val);
+		free(c->memory);
+	}
 
 	free(clients->client);
 	free(clients);
@@ -2012,6 +2038,9 @@ print_clients_header(struct igt_drm_clients *clients, int lines,
 		if (lines++ >= con_h || len >= con_w)
 			return lines;
 
+		if (iclients->regions)
+			len += printf("     MEM      RSS ");
+
 		if (iclients->classes.num_engines) {
 			unsigned int i;
 			int width;
@@ -2055,6 +2084,20 @@ print_clients_header(struct igt_drm_clients *clients, int lines,
 static bool numeric_clients;
 static bool filter_idle;
 
+static int print_size(uint64_t sz)
+{
+	char units[] = { ' ', 'K', 'M', 'G' };
+	unsigned int u;
+
+	for (u = 0; u < ARRAY_SIZE(units) - 1; u++) {
+		if (sz & 1023 || sz < 1024)
+			break;
+		sz /= 1024;
+	}
+
+	return printf("%7"PRIu64"%c ", sz, units[u]);
+}
+
 static int
 print_client(struct igt_drm_client *c, struct engines *engines, double t, int lines,
 	     int con_w, int con_h, unsigned int period_us, int *class_w)
@@ -2072,6 +2115,18 @@ print_client(struct igt_drm_client *c, struct engines *engines, double t, int li
 
 		len = printf("%*s ", clients->max_pid_len, c->pid_str);
 
+		if (iclients->regions) {
+			uint64_t sz;
+
+			for (sz = 0, i = 0; i <= c->regions->max_region_id; i++)
+				sz += c->memory[i].total;
+			len += print_size(sz);
+
+			for (sz = 0, i = 0; i <= c->regions->max_region_id; i++)
+				sz += c->memory[i].resident;
+			len += print_size(sz);
+		}
+
 		for (i = 0; i <= iclients->classes.max_engine_id; i++) {
 			double pct, max;
 
@@ -2111,6 +2166,42 @@ print_client(struct igt_drm_client *c, struct engines *engines, double t, int li
 		snprintf(buf, sizeof(buf), "%u", c->pid);
 		__json_add_member("pid", buf);
 
+		if (iclients->regions) {
+			pops->open_struct("memory");
+
+			for (i = 0; i < ARRAY_SIZE(json_memory_region_names);
+			     i++) {
+				if (i > c->regions->max_region_id)
+					break;
+
+				pops->open_struct(json_memory_region_names[i]);
+
+				snprintf(buf, sizeof(buf), "%" PRIu64,
+					 c->memory[i].total);
+				__json_add_member("total", buf);
+
+				snprintf(buf, sizeof(buf), "%" PRIu64,
+					 c->memory[i].shared);
+				__json_add_member("shared", buf);
+
+				snprintf(buf, sizeof(buf), "%" PRIu64,
+					 c->memory[i].resident);
+				__json_add_member("resident", buf);
+
+				snprintf(buf, sizeof(buf), "%" PRIu64,
+					 c->memory[i].purgeable);
+				__json_add_member("purgeable", buf);
+
+				snprintf(buf, sizeof(buf), "%" PRIu64,
+					 c->memory[i].active);
+				__json_add_member("active", buf);
+
+				pops->close_struct();
+			}
+
+			pops->close_struct();
+		}
+
 		if (c->samples > 1) {
 			pops->open_struct("engine-classes");
 
@@ -2456,10 +2547,29 @@ static void intel_scan_clients(struct intel_clients *iclients)
 		"video-enhance",
 		"compute",
 	};
+	struct igt_drm_client *c;
+	unsigned int i;
 
 	igt_drm_clients_scan(iclients->clients, client_match,
 			     engine_map, ARRAY_SIZE(engine_map),
-			     NULL, 0);
+			     memory_region_map, ARRAY_SIZE(memory_region_map));
+
+	iclients->regions = NULL;
+
+	if (!iclients->clients)
+		return;
+
+	/*
+	 * Borrow memory region data from first client so we can use it
+	 * when displaying stuff. All regions are the same due our client_match.
+	 */
+	igt_for_each_drm_client(iclients->clients, c, i) {
+		if (c->status == IGT_DRM_CLIENT_ALIVE) {
+			if (c->regions->num_regions)
+				iclients->regions = c->regions;
+			break;
+		}
+	}
 }
 
 int main(int argc, char **argv)
-- 
2.39.2

[Intel-gfx] [PATCH i-g-t 12/12] tools/intel_gpu_top: Add ability to show memory region breakdown

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Similar as we can toggle between aggregated engines and clients, add the
capability to toggle between aggregated and per memory region stats.

It starts in aggregated mode by default and interactive command 'm' and
command line switch '-m' can be used to toggle that.

Both only affect the interactive view, while JSON output always contains
separate memory regions.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 man/intel_gpu_top.rst |  4 ++++
 tools/intel_gpu_top.c | 56 +++++++++++++++++++++++++++++++++----------
 2 files changed, 48 insertions(+), 12 deletions(-)

diff --git a/man/intel_gpu_top.rst b/man/intel_gpu_top.rst
index 9499f87f1b4d..44a54a5f219d 100644
--- a/man/intel_gpu_top.rst
+++ b/man/intel_gpu_top.rst
@@ -55,6 +55,9 @@ OPTIONS
 -p
    Default to showing physical engines instead of aggregated classes.
 
+-m
+   Default to showing all memory regions separately.
+
 RUNTIME CONTROL
 ===============
 
@@ -68,6 +71,7 @@ Supported keys:
 |    's'    Toggle between sort modes (runtime, total runtime, pid, client id).
 |    'i'    Toggle display of clients which used no GPU time.
 |    'H'    Toggle between per PID aggregation and individual clients.
+|    'm'    Toggle between aggregated memory regions and full breakdown.
 
 DEVICE SELECTION
 ================
diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 958df42e8c20..3f376e233fe2 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -138,6 +138,7 @@ struct intel_clients {
 
 static struct termios termios_orig;
 static bool class_view;
+static bool aggregate_regions;
 
 /* Maps i915 fdinfo names to indices */
 static const char *memory_region_map[] = {
@@ -1045,6 +1046,7 @@ usage(const char *appname)
 		"\t[-L]            List all cards.\n"
 		"\t[-d <device>]   Device filter, please check manual page for more details.\n"
 		"\t[-p]            Default to showing physical engines instead of classes.\n"
+		"\t[-m]            Default to showing all memory regions.\n"
 		"\n",
 		appname, DEFAULT_PERIOD_MS);
 	igt_device_print_filter_types();
@@ -2026,7 +2028,7 @@ print_clients_header(struct igt_drm_clients *clients, int lines,
 				 4 : clients->max_name_len; /* At least "NAME" */
 
 	if (output_mode == INTERACTIVE) {
-		unsigned int num_active = 0;
+		unsigned int num_active = 0, i;
 		int len;
 
 		if (lines++ >= con_h)
@@ -2038,11 +2040,17 @@ print_clients_header(struct igt_drm_clients *clients, int lines,
 		if (lines++ >= con_h || len >= con_w)
 			return lines;
 
-		if (iclients->regions)
-			len += printf("     MEM      RSS ");
+		if (iclients->regions) {
+			if (aggregate_regions) {
+				len += printf("     MEM      RSS ");
+			} else {
+				len += printf("     RAM      RSS ");
+				if (iclients->regions->num_regions > 1)
+					len += printf("    VRAM     VRSS ");
+			}
+		}
 
 		if (iclients->classes.num_engines) {
-			unsigned int i;
 			int width;
 
 			for (i = 0; i <= iclients->classes.max_engine_id; i++) {
@@ -2116,15 +2124,26 @@ print_client(struct igt_drm_client *c, struct engines *engines, double t, int li
 		len = printf("%*s ", clients->max_pid_len, c->pid_str);
 
 		if (iclients->regions) {
-			uint64_t sz;
+			if (aggregate_regions) {
+				uint64_t sz;
 
-			for (sz = 0, i = 0; i <= c->regions->max_region_id; i++)
-				sz += c->memory[i].total;
-			len += print_size(sz);
+				for (sz = 0, i = 0;
+				     i <= c->regions->max_region_id; i++)
+					sz += c->memory[i].total;
+				len += print_size(sz);
 
-			for (sz = 0, i = 0; i <= c->regions->max_region_id; i++)
-				sz += c->memory[i].resident;
-			len += print_size(sz);
+				for (sz = 0, i = 0;
+				     i <= c->regions->max_region_id; i++)
+					sz += c->memory[i].resident;
+				len += print_size(sz);
+			} else {
+				len += print_size(c->memory[0].total);
+				len += print_size(c->memory[0].resident);
+				if (c->regions->num_regions > 1) {
+					len += print_size(c->memory[1].total);
+					len += print_size(c->memory[1].resident);
+				}
+			}
 		}
 
 		for (i = 0; i <= iclients->classes.max_engine_id; i++) {
@@ -2401,6 +2420,13 @@ static void process_normal_stdin(void)
 			else
 				header_msg = "Showing individual clients.";
 			break;
+		case 'm':
+			aggregate_regions ^= true;
+			if (aggregate_regions)
+				header_msg = "Aggregating memory regions.";
+			else
+				header_msg = "Showing memory regions.";
+			break;
 		};
 	}
 }
@@ -2449,6 +2475,7 @@ static void show_help_screen(void)
 "    's'    Toggle between sort modes (runtime, total runtime, pid, client id).\n"
 "    'i'    Toggle display of clients which used no GPU time.\n"
 "    'H'    Toggle between per PID aggregation and individual clients.\n"
+"    'm'    Toggle between aggregated memory regions and full breakdown.\n"
 "\n"
 "    'h' or 'q'    Exit interactive help.\n"
 "\n");
@@ -2576,6 +2603,7 @@ int main(int argc, char **argv)
 {
 	unsigned int period_us = DEFAULT_PERIOD_MS * 1000;
 	bool physical_engines = false;
+	bool separate_regions = false;
 	struct intel_clients iclients;
 	int con_w = -1, con_h = -1;
 	char *output_path = NULL;
@@ -2588,7 +2616,7 @@ int main(int argc, char **argv)
 	struct timespec ts;
 
 	/* Parse options */
-	while ((ch = getopt(argc, argv, "o:s:d:pcJLlh")) != -1) {
+	while ((ch = getopt(argc, argv, "o:s:d:mpcJLlh")) != -1) {
 		switch (ch) {
 		case 'o':
 			output_path = optarg;
@@ -2602,6 +2630,9 @@ int main(int argc, char **argv)
 		case 'p':
 			physical_engines = true;
 			break;
+		case 'm':
+			separate_regions = true;
+			break;
 		case 'c':
 			output_mode = CSV;
 			break;
@@ -2645,6 +2676,7 @@ int main(int argc, char **argv)
 		fprintf(stderr, "Failed to install signal handler!\n");
 
 	class_view = !physical_engines;
+	aggregate_regions = !separate_regions;
 
 	switch (output_mode) {
 	case INTERACTIVE:
-- 
2.39.2