tools/intel_perf_counters.c

/*
 * Copyright © 2010, 2013 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, 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:
 *    Eric Anholt <eric@anholt.net>
 *    Kenneth Graunke <kenneth@whitecape.org>
 *
 * While documentation for performance counters is suspiciously missing from the
 * Sandybridge PRM, they were documented in Volume 1 Part 3 of the Ironlake PRM.
 *
 * A lot of the Ironlake PRM actually unintentionally documents Sandybridge
 * due to mistakes made when updating the documentation for Gen6+.  Many of
 * these mislabeled sections carried forward to the public documentation.
 *
 * The Ironlake PRMs have been publicly available since 2010 and are online at:
 * https://01.org/linuxgraphics/documentation/2010-intel-core-processor-family
 */

#include <unistd.h>
#include <stdbool.h>
#include <stdlib.h>
#include <stdio.h>
#include <err.h>
#include <sys/ioctl.h>

#include "drm.h"
#include "i915_drm.h"
#include "drmtest.h"
#include "intel_gpu_tools.h"
#include "intel_bufmgr.h"
#include "intel_batchbuffer.h"

#define GEN5_COUNTER_COUNT 29

const char *gen5_counter_names[GEN5_COUNTER_COUNT] = {
	"cycles the CS unit is starved",
	"cycles the CS unit is stalled",
	"cycles the VF unit is starved",
	"cycles the VF unit is stalled",
	"cycles the VS unit is starved",
	"cycles the VS unit is stalled",
	"cycles the GS unit is starved",
	"cycles the GS unit is stalled",
	"cycles the CL unit is starved",
	"cycles the CL unit is stalled",
	"cycles the SF unit is starved",
	"cycles the SF unit is stalled",
	"cycles the WZ unit is starved",
	"cycles the WZ unit is stalled",
	"Z buffer read/write          ",
	"cycles each EU was active    ",
	"cycles each EU was suspended ",
	"cycles threads loaded all EUs",
	"cycles filtering active      ",
	"cycles PS threads executed   ",
	"subspans written to RC       ",
	"bytes read for texture reads ",
	"texels returned from sampler ",
	"polygons not culled          ",
	"clocks MASF has valid message",
	"64b writes/reads from RC     ",
	"reads on dataport            ",
	"clocks MASF has valid msg not consumed by sampler",
	"cycles any EU is stalled for math",
};

#define GEN6_COUNTER_COUNT 29

/**
 * Sandybridge: Counter Select = 001
 * A0   A1   A2   A3   A4   TIMESTAMP RPT_ID
 * A5   A6   A7   A8   A9   A10  A11  A12
 * A13  A14  A15  A16  A17  A18  A19  A20
 * A21  A22  A23  A24  A25  A26  A27  A28
 */
const int gen6_counter_format = 1;

/**
 * Names for aggregating counters A0-A28.
 *
 * While the Ironlake PRM clearly documents that there are 29 counters (A0-A28),
 * it only lists the names for 28 of them; one is missing.  However, careful
 * examination reveals a pattern: there are five GS counters (Active, Stall,
 * Core Stall, # threads loaded, and ready but not running time).  There are
 * also five PS counters, in the same order.  But there are only four VS
 * counters listed - the number of VS threads loaded is missing.  Presumably,
 * it exists and is counter 5, and the rest are shifted over one place.
 */
const char *gen6_counter_names[GEN6_COUNTER_COUNT] = {
	[0]  = "Aggregated Core Array Active",
	[1]  = "Aggregated Core Array Stalled",
	[2]  = "Vertex Shader Active Time",
	[3]  = "Vertex Shader Stall Time",
	[4]  = "Vertex Shader Stall Time - Core Stall",
	[5]  = "# VS threads loaded",
	[6]  = "Vertex Shader Ready but not running time",
	[7]  = "Geometry Shader Active Time",
	[8]  = "Geometry Shader Stall Time",
	[9]  = "Geometry Shader Stall Time - Core Stall",
	[10] = "# GS threads loaded",
	[11] = "Geometry Shader ready but not running Time",
	[12] = "Pixel Shader Active Time",
	[13] = "Pixel Shader Stall Time",
	[14] = "Pixel Shader Stall Time - Core Stall",
	[15] = "# PS threads loaded",
	[16] = "Pixel Shader ready but not running Time",
	[17] = "Early Z Test Pixels Passing",
	[18] = "Early Z Test Pixels Failing",
	[19] = "Early Stencil Test Pixels Passing",
	[20] = "Early Stencil Test Pixels Failing",
	[21] = "Pixel Kill Count",
	[22] = "Alpha Test Pixels Failed",
	[23] = "Post PS Stencil Pixels Failed",
	[24] = "Post PS Z buffer Pixels Failed",
	[25] = "Pixels/samples Written in the frame buffer",
	[26] = "GPU Busy",
	[27] = "CL active and not stalled",
	[28] = "SF active and stalled",
};

#define GEN7_COUNTER_COUNT 44

/**
 * Names for aggregating counters A0-A44.  Uninitialized fields are "Reserved."
 */
const char *gen7_counter_names[GEN7_COUNTER_COUNT] = {
	/* A0:
	 * The sum of all cycles on all cores actively executing instructions
	 * This does not count the time taken to service Send instructions.
	 * This time is considered by shader active counters to give the result.
	 */
	[0]  = "Aggregated Core Array Active",
	/* A1:
	 * The sum of all cycles on all cores where the EU is not idle and is
	 * not actively executing ISA instructions.  Generally this means that
	 * all loaded threads on the EU are stalled on some data dependency,
	 * but this also includes the time during which the TS is loading the
	 * thread dispatch header into the EU prior to thread execution and no
	 * other thread is fully loaded.
	 */
	[1]  = "Aggregated Core Array Stalled",
	/* A2:
	 * Total time in clocks the vertex shader spent active on all cores.
	 */
	[2]  = "Vertex Shader Active Time",
	/* A4:
	 * Total time in clocks the vertex shader spent stalled on all cores -
	 * and the entire core was stalled as well.
	 */
	[4]  = "Vertex Shader Stall Time - Core Stall",
	/* A5: Number of VS threads loaded at any given time in the EUs. */
	[5]  = "# VS threads loaded",
	/* A7:
	 * Total time in clocks the Hull shader spent active on all cores.
	 */
	[7]  = "Hull Shader Active Time",
	/* A9:
	 * Total time in clocks the Hull shader spent stalled on all cores -
	 * and the entire core was stalled as well.
	 */
	[9]  = "Hull Shader Stall Time - Core Stall",
	/* A10: Number of HS threads loaded at any given time in the EUs. */
	[10] = "# HS threads loaded",
	/* A12:
	 * Total time in clocks the Domain shader spent active on all cores.
	 */
	[12] = "Domain Shader Active Time",
	/* A14:
	 * Total time in clocks the domain shader spent stalled on all cores -
	 * and the entire core was stalled as well.
	 */
	[14] = "Domain Shader Stall Time - Core Stall",
	/* A15: Number of DS threads loaded at any given time in the EUs. */
	[15] = "# DS threads loaded",
	/* A17:
	 * Total time in clocks the compute shader spent active on all cores.
	 */
	[17] = "Compute Shader Active Time",
	/* A19:
	 * Total time in clocks the compute shader spent stalled on all cores -
	 * and the entire core was stalled as well.
	 */
	[19] = "Compute Shader Stall Time - Core Stall",
	/* A20: Number of CS threads loaded at any given time in the EUs. */
	[20] = "# CS threads loaded",
	/* A22:
	 * Total time in clocks the geometry shader spent active on all cores.
	 */
	[22] = "Geometry Shader Active Time",
	/* A24:
	 * Total time in clocks the geometry shader spent stalled on all cores -
	 * and the entire core was stalled as well.
	 */
	[24] = "Geometry Shader Stall Time - Core Stall",
	/* A25: Number of GS threads loaded at any time in the EUs. */
	[25] = "# GS threads loaded",
	/* A27:
	 * Total time in clocks the pixel shader spent active on all cores.
	 */
	[27] = "Pixel Shader Active Time",
	/* A29:
	 * Total time in clocks the pixel shader spent stalled on all cores -
	 * and the entire core was stalled as well.
	 */
	[29] = "Pixel Shader Stall Time - Core Stall",
	/* A30: Number of PS threads loaded at any given time in the EUs. */
	[30] = "# PS threads loaded",
	/* A32: Count of pixels that pass the fast check (8x8). */
	[32] = "HiZ Fast Z Test Pixels Passing",
	/* A33: Count of pixels that fail the fast check (8x8). */
	[33] = "HiZ Fast Z Test Pixels Failing",
	/* A34: Count of pixels passing the slow check (2x2). */
	[34] = "Slow Z Test Pixels Passing",
	/* A35: Count of pixels that fail the slow check (2x2). */
	[35] = "Slow Z Test Pixels Failing",
	/* A36: Number of pixels/samples killed in the pixel shader.
	 * Ivybridge/Baytrail Erratum: Count reported is 2X the actual count for
	 * dual source render target messages i.e. when PS has two output colors.
	 */
	[36] = "Pixel Kill Count",
	/* A37:
	 * Number of pixels/samples that fail alpha-test.  Alpha to coverage
	 * may have some challenges in per-pixel invocation.
	 */
	[37] = "Alpha Test Pixels Failed",
	/* A38:
	 * Number of pixels/samples failing stencil test after the pixel shader
	 * has executed.
	 */
	[38] = "Post PS Stencil Pixels Failed",
	/* A39:
	 * Number of pixels/samples fail Z test after the pixel shader has
	 * executed.
	 */
	[39] = "Post PS Z buffer Pixels Failed",
	/* A40:
	 * Number of render target writes.  MRT scenarios will cause this
	 * counter to increment multiple times.
	 */
	[40] = "3D/GPGPU Render Target Writes",
	/* A41: Render engine is not idle.
	 *
	 * GPU Busy aggregate counter doesn't increment under the following
	 * conditions:
	 *
	 * 1. Context Switch in Progress.
	 * 2. GPU stalled on executing MI_WAIT_FOR_EVENT.
	 * 3. GPU stalled on execution MI_SEMAPHORE_MBOX.
	 * 4. RCS idle but other parts of GPU active (e.g. only media engines
	 *    active)
	 */
	[41] = "Render Engine Busy",
	/* A42:
	 * VSunit is stalling VF (upstream unit) and starving HS (downstream
	 * unit).
	 */
	[42] = "VS bottleneck",
	/* A43:
	 * GSunit is stalling DS (upstream unit) and starving SOL (downstream
	 * unit).
	 */
	[43] = "GS bottleneck",
};

/**
 * Ivybridge - Counter Select = 101
 * A4   A3   A2   A1   A0   TIMESTAMP  ReportID
 * A12  A11  A10  A9   A8   A7   A6    A5
 * A20  A19  A18  A17  A16  A15  A14   A13
 * A28  A27  A26  A25  A24  A23  A22   A21
 * A36  A35  A34  A33  A32  A31  A30   A29
 * A44  A43  A42  A41  A40  A39  A38   A37
 * C3   C2   C1   C0   B3   B2   B1    B0
 * C11  C10  C9   C8   C7   C6   C5    C4
 */
const int gen7_counter_format = 5; /* 0b101 */

int have_totals = 0;
uint32_t *totals;
uint32_t *last_counter;
static drm_intel_bufmgr *bufmgr;
struct intel_batchbuffer *batch;

/* DW0 */
#define GEN5_MI_REPORT_PERF_COUNT ((0x26 << 23) | (3 - 2))
#define MI_COUNTER_SET_0	(0 << 6)
#define MI_COUNTER_SET_1	(1 << 6)
/* DW1 */
#define MI_COUNTER_ADDRESS_GTT	(1 << 0)
/* DW2: report ID */

/**
 * According to the Sandybridge PRM, Volume 1, Part 1, page 48,
 * MI_REPORT_PERF_COUNT is now opcode 0x28.  The Ironlake PRM, Volume 1,
 * Part 3 details how it works.
 */
/* DW0 */
#define GEN6_MI_REPORT_PERF_COUNT (0x28 << 23)
/* DW1 and 2 are the same as above */

/* OACONTROL exists on Gen6+ but is documented in the Ironlake PRM */
#define OACONTROL                       0x2360
# define OACONTROL_COUNTER_SELECT_SHIFT 2
# define PERFORMANCE_COUNTER_ENABLE     (1 << 0)

static void
gen5_get_counters(void)
{
	int i;
	drm_intel_bo *stats_bo;
	uint32_t *stats_result;

	stats_bo = drm_intel_bo_alloc(bufmgr, "stats", 4096, 4096);

	BEGIN_BATCH(6);
	OUT_BATCH(GEN5_MI_REPORT_PERF_COUNT | MI_COUNTER_SET_0);
	OUT_RELOC(stats_bo,
		  I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
		  0);
	OUT_BATCH(0);

	OUT_BATCH(GEN5_MI_REPORT_PERF_COUNT | MI_COUNTER_SET_1);
	OUT_RELOC(stats_bo,
		  I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
		  64);
	OUT_BATCH(0);

	ADVANCE_BATCH();

	intel_batchbuffer_flush(batch);

	drm_intel_bo_map(stats_bo, 0);
	stats_result = stats_bo->virtual;
	/* skip REPORT_ID, TIMESTAMP */
	stats_result += 3;
	for (i = 0 ; i < GEN5_COUNTER_COUNT; i++) {
		totals[i] += stats_result[i] - last_counter[i];
		last_counter[i] = stats_result[i];
	}

	drm_intel_bo_unmap(stats_bo);
	drm_intel_bo_unreference(stats_bo);
}

static void
gen6_get_counters(void)
{
	int i;
	drm_intel_bo *stats_bo;
	uint32_t *stats_result;

	/* Map from counter names to their index in the buffer object */
	static const int buffer_index[GEN6_COUNTER_COUNT] =
	{
		7,   6,  5,  4,  3,
		15, 14, 13, 12, 11, 10,  9,  8,
		23, 22, 21, 20, 19, 18, 17, 16,
		31, 30, 29, 28, 27, 26, 25, 24,
	};

	stats_bo = drm_intel_bo_alloc(bufmgr, "stats", 4096, 4096);

	BEGIN_BATCH(3);
	OUT_BATCH(GEN6_MI_REPORT_PERF_COUNT | (3 - 2));
	OUT_RELOC(stats_bo,
		  I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
		  MI_COUNTER_ADDRESS_GTT);
	OUT_BATCH(0);
	ADVANCE_BATCH();

	intel_batchbuffer_flush_on_ring(batch, I915_EXEC_RENDER);

	drm_intel_bo_map(stats_bo, 0);
	stats_result = stats_bo->virtual;
	for (i = 0; i < GEN6_COUNTER_COUNT; i++) {
		totals[i] += stats_result[buffer_index[i]] - last_counter[i];
		last_counter[i] = stats_result[buffer_index[i]];
	}

	drm_intel_bo_unmap(stats_bo);
	drm_intel_bo_unreference(stats_bo);
}

static void
gen7_get_counters(void)
{
	int i;
	drm_intel_bo *stats_bo;
	uint32_t *stats_result;

	stats_bo = drm_intel_bo_alloc(bufmgr, "stats", 4096, 4096);

	BEGIN_BATCH(3);
	OUT_BATCH(GEN6_MI_REPORT_PERF_COUNT | (3 - 2));
	OUT_RELOC(stats_bo,
		  I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, 0);
	OUT_BATCH(0);
	ADVANCE_BATCH();

	intel_batchbuffer_flush_on_ring(batch, I915_EXEC_RENDER);

	drm_intel_bo_map(stats_bo, 0);
	stats_result = stats_bo->virtual;
	/* skip REPORT_ID, TIMESTAMP */
	stats_result += 3;
	for (i = 0; i < GEN7_COUNTER_COUNT; i++) {
		/* Ignore "Reserved" counters */
		if (!gen7_counter_names[i])
			continue;
		totals[i] += stats_result[i] - last_counter[i];
		last_counter[i] = stats_result[i];
	}

	drm_intel_bo_unmap(stats_bo);
	drm_intel_bo_unreference(stats_bo);
}

#define STATS_CHECK_FREQUENCY	100
#define STATS_REPORT_FREQUENCY	2

int
main(int argc, char **argv)
{
	uint32_t devid;
	int counter_format;
	int counter_count;
	const char **counter_name;
	void (*get_counters)(void);
	int i;
	char clear_screen[] = {0x1b, '[', 'H',
			       0x1b, '[', 'J',
			       0x0};
	bool oacontrol = true;
	int fd;
	int l;

	fd = drm_open_any();
	devid = intel_get_drm_devid(fd);

	bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
	drm_intel_bufmgr_gem_enable_reuse(bufmgr);
	batch = intel_batchbuffer_alloc(bufmgr, devid);

	if (IS_GEN5(devid)) {
		counter_name = gen5_counter_names;
		counter_count = GEN5_COUNTER_COUNT;
		get_counters = gen5_get_counters;
		oacontrol = false;
	} else if (IS_GEN6(devid)) {
		counter_name = gen6_counter_names;
		counter_count = GEN6_COUNTER_COUNT;
		counter_format = gen6_counter_format;
		get_counters = gen6_get_counters;
	} else if (IS_GEN7(devid)) {
		counter_name = gen7_counter_names;
		counter_count = GEN7_COUNTER_COUNT;
		counter_format = gen7_counter_format;
		get_counters = gen7_get_counters;
	} else {
		printf("This tool is not yet supported on your platform.\n");
		abort();
	}

	if (oacontrol) {
		/* Forcewake */
		intel_register_access_init(intel_get_pci_device(), 0);

		/* Enable performance counters */
		intel_register_write(OACONTROL,
			counter_format << OACONTROL_COUNTER_SELECT_SHIFT |
			PERFORMANCE_COUNTER_ENABLE);
	}

	totals = calloc(counter_count, sizeof(uint32_t));
	last_counter = calloc(counter_count, sizeof(uint32_t));

	for (;;) {
		for (l = 0; l < STATS_CHECK_FREQUENCY; l++) {
			printf("%s", clear_screen);

			if (l % (STATS_CHECK_FREQUENCY / STATS_REPORT_FREQUENCY) == 0) {
				if (have_totals) {
					for (i = 0; i < counter_count; i++) {
						/* Ignore "Reserved" counters */
						if (!counter_name[i])
							continue;
						printf("%s: %u\n", counter_name[i],
						       totals[i]);
						totals[i] = 0;
					}
				}
			}

			get_counters();
			have_totals = 1;

			usleep(1000000 / STATS_CHECK_FREQUENCY);
		}
	}

	if (oacontrol) {
		/* Disable performance counters */
		intel_register_write(OACONTROL, 0);

		/* Forcewake */
		intel_register_access_fini();
	}

	free(totals);
	free(last_counter);

	return 0;
}