* [igt-dev] [PATCH i-g-t 3/6] lib/panfrost: Add panfrost helpers
2019-05-10 14:44 [igt-dev] [PATCH i-g-t 1/6] lib: Check for -ETIME, not ETIME Tomeu Vizoso
2019-05-10 14:44 ` [igt-dev] [PATCH i-g-t 2/6] drm-uapi: Add panfrost header Tomeu Vizoso
@ 2019-05-10 14:44 ` Tomeu Vizoso
2019-05-10 14:44 ` [igt-dev] [PATCH i-g-t 4/6] lib: Add support for opening panfrost devices Tomeu Vizoso
` (4 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Tomeu Vizoso @ 2019-05-10 14:44 UTC (permalink / raw)
To: Development mailing list for IGT GPU Tools; +Cc: Petri Latvala, Tomeu Vizoso
I also needed to copy two headers from Mesa so we have the struct
definitions involved in a trivial clear job. These headers were
unchanged.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Acked-by: Petri Latvala <petri.latvala@intel.com>
---
lib/igt_panfrost.c | 244 +++++++
lib/igt_panfrost.h | 60 ++
lib/meson.build | 1 +
lib/panfrost-job.h | 1516 +++++++++++++++++++++++++++++++++++++++++++
lib/panfrost-misc.h | 47 ++
5 files changed, 1868 insertions(+)
create mode 100644 lib/igt_panfrost.c
create mode 100644 lib/igt_panfrost.h
create mode 100644 lib/panfrost-job.h
create mode 100644 lib/panfrost-misc.h
diff --git a/lib/igt_panfrost.c b/lib/igt_panfrost.c
new file mode 100644
index 000000000000..1046c8260552
--- /dev/null
+++ b/lib/igt_panfrost.c
@@ -0,0 +1,244 @@
+/*
+ * Copyright © 2016 Broadcom
+ * Copyright © 2019 Collabora, Ltd.
+ *
+ * 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.
+ */
+
+#include <assert.h>
+#include <string.h>
+#include <signal.h>
+#include <errno.h>
+#include <sys/mman.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/ioctl.h>
+#include <fcntl.h>
+
+#include "drmtest.h"
+#include "igt_aux.h"
+#include "igt_core.h"
+#include "igt_panfrost.h"
+#include "ioctl_wrappers.h"
+#include "intel_reg.h"
+#include "intel_chipset.h"
+#include "panfrost_drm.h"
+#include "panfrost-job.h"
+
+/**
+ * SECTION:igt_panfrost
+ * @short_description: PANFROST support library
+ * @title: PANFROST
+ * @include: igt.h
+ *
+ * This library provides various auxiliary helper functions for writing PANFROST
+ * tests.
+ */
+
+struct panfrost_bo *
+igt_panfrost_gem_new(int fd, size_t size)
+{
+ struct panfrost_bo *bo = calloc(1, sizeof(*bo));
+
+ struct drm_panfrost_create_bo create_bo = {
+ .size = size,
+ };
+
+ do_ioctl(fd, DRM_IOCTL_PANFROST_CREATE_BO, &create_bo);
+
+ bo->handle = create_bo.handle;
+ bo->offset = create_bo.offset;
+ bo->size = size;
+ return bo;
+}
+
+void
+igt_panfrost_free_bo(int fd, struct panfrost_bo *bo)
+{
+ if (bo->map)
+ munmap(bo->map, bo->size);
+ gem_close(fd, bo->handle);
+ free(bo);
+}
+
+uint32_t
+igt_panfrost_get_bo_offset(int fd, uint32_t handle)
+{
+ struct drm_panfrost_get_bo_offset get = {
+ .handle = handle,
+ };
+
+ do_ioctl(fd, DRM_IOCTL_PANFROST_GET_BO_OFFSET, &get);
+
+ return get.offset;
+}
+
+uint32_t
+igt_panfrost_get_param(int fd, int param)
+{
+ struct drm_panfrost_get_param get = {
+ .param = param,
+ };
+
+ do_ioctl(fd, DRM_IOCTL_PANFROST_GET_PARAM, &get);
+
+ return get.value;
+}
+
+void *
+igt_panfrost_mmap_bo(int fd, uint32_t handle, uint32_t size, unsigned prot)
+{
+ struct drm_panfrost_mmap_bo mmap_bo = {
+ .handle = handle,
+ };
+ void *ptr;
+
+ mmap_bo.handle = handle;
+ do_ioctl(fd, DRM_IOCTL_PANFROST_MMAP_BO, &mmap_bo);
+
+ ptr = mmap(0, size, prot, MAP_SHARED, fd, mmap_bo.offset);
+ if (ptr == MAP_FAILED)
+ return NULL;
+ else
+ return ptr;
+}
+
+void igt_panfrost_bo_mmap(int fd, struct panfrost_bo *bo)
+{
+ bo->map = igt_panfrost_mmap_bo(fd, bo->handle, bo->size,
+ PROT_READ | PROT_WRITE);
+ igt_assert(bo->map);
+}
+
+/* TODO: Make this more generic so it works on GPUs other than T760 */
+struct panfrost_submit *igt_panfrost_trivial_job(int fd, bool do_crash, int width, int height, uint32_t color)
+{
+ struct panfrost_submit *submit;
+ struct mali_job_descriptor_header header = {
+ .job_type = JOB_TYPE_FRAGMENT,
+ .job_index = 1,
+#ifdef __LP64__
+ .job_descriptor_size = 1,
+#endif
+ };
+ struct mali_payload_fragment payload = {
+ .min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(0, 0),
+ .max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(ALIGN(width, 16), height),
+ };
+ struct bifrost_framebuffer framebuffer = {
+ .unk0 = 0x0,
+ .unknown1 = 0x0,
+ .tiler_meta = 0xff00000000,
+ .width1 = MALI_POSITIVE(ALIGN(width, 16)),
+ .height1 = MALI_POSITIVE(height),
+ .width2 = MALI_POSITIVE(ALIGN(width, 16)),
+ .height2 = MALI_POSITIVE(height),
+ .unk1 = 0x1080,
+ .unk2 = 0x0,
+ .rt_count_1 = MALI_POSITIVE(1),
+ .rt_count_2 = 1,
+ .unk3 = 0x100,
+ .clear_stencil = 0x0,
+ .clear_depth = 0.000000,
+ .unknown2 = 0x1f,
+ };
+ struct mali_rt_format fmt = {
+ .unk1 = 0x4000000,
+ .unk2 = 0x1,
+ .nr_channels = MALI_POSITIVE(4),
+ .flags = do_crash ? 0x444 | (1 << 8) : 0x444,
+ .swizzle = MALI_CHANNEL_BLUE | (MALI_CHANNEL_GREEN << 3) | (MALI_CHANNEL_RED << 6) | (MALI_CHANNEL_ONE << 9),
+ .unk4 = 0x8,
+ };
+ struct bifrost_render_target rts = {
+ .format = fmt,
+ .chunknown = {
+ .unk = 0x0,
+ .pointer = 0x0,
+ },
+ .framebuffer_stride = ALIGN(width, 16) * 4 / 16,
+ .clear_color_1 = color,
+ .clear_color_2 = color,
+ .clear_color_3 = color,
+ .clear_color_4 = color,
+ };
+ uint32_t *bos;
+
+ submit = malloc(sizeof(*submit));
+
+ submit->fbo = igt_panfrost_gem_new(fd, ALIGN(width, 16) * height * 4);
+ rts.framebuffer = submit->fbo->offset;
+
+ submit->tiler_heap_bo = igt_panfrost_gem_new(fd, 32768 * 4096);
+ framebuffer.tiler_heap_start = submit->tiler_heap_bo->offset;
+ framebuffer.tiler_heap_end = submit->tiler_heap_bo->offset + 32768 * 4096;
+
+ submit->tiler_scratch_bo = igt_panfrost_gem_new(fd, 128 * 128 * 4096);
+ framebuffer.tiler_scratch_start = submit->tiler_scratch_bo->offset;
+ framebuffer.tiler_scratch_middle = submit->tiler_scratch_bo->offset + 0xf0000;
+
+ submit->scratchpad_bo = igt_panfrost_gem_new(fd, 64 * 4096);
+ framebuffer.scratchpad = submit->scratchpad_bo->offset;
+
+ submit->fb_bo = igt_panfrost_gem_new(fd, sizeof(framebuffer) + sizeof(struct bifrost_render_target));
+ igt_panfrost_bo_mmap(fd, submit->fb_bo);
+ memcpy(submit->fb_bo->map, &framebuffer, sizeof(framebuffer));
+ memcpy(submit->fb_bo->map + sizeof(framebuffer), &rts, sizeof(struct bifrost_render_target));
+ payload.framebuffer = submit->fb_bo->offset | MALI_MFBD;
+
+ submit->submit_bo = igt_panfrost_gem_new(fd, sizeof(header) + sizeof(payload));
+ igt_panfrost_bo_mmap(fd, submit->submit_bo);
+
+ memcpy(submit->submit_bo->map, &header, sizeof(header));
+ memcpy(submit->submit_bo->map + sizeof(header), &payload, sizeof(payload));
+
+ submit->args = malloc(sizeof(*submit->args));
+ memset(submit->args, 0, sizeof(*submit->args));
+ submit->args->jc = submit->submit_bo->offset;
+ submit->args->requirements = PANFROST_JD_REQ_FS;
+
+ bos = malloc(sizeof(*bos) * 6);
+ bos[0] = submit->fbo->handle;
+ bos[1] = submit->tiler_heap_bo->handle;
+ bos[2] = submit->tiler_scratch_bo->handle;
+ bos[3] = submit->scratchpad_bo->handle;
+ bos[4] = submit->fb_bo->handle;
+ bos[5] = submit->submit_bo->handle;
+
+ submit->args->bo_handles = to_user_pointer(bos);
+ submit->args->bo_handle_count = 6;
+
+ igt_assert_eq(drmSyncobjCreate(fd, DRM_SYNCOBJ_CREATE_SIGNALED, &submit->args->out_sync), 0);
+
+ return submit;
+}
+
+void igt_panfrost_free_job(int fd, struct panfrost_submit *submit)
+{
+ free(from_user_pointer(submit->args->bo_handles));
+ igt_panfrost_free_bo(fd, submit->submit_bo);
+ igt_panfrost_free_bo(fd, submit->fb_bo);
+ igt_panfrost_free_bo(fd, submit->scratchpad_bo);
+ igt_panfrost_free_bo(fd, submit->tiler_scratch_bo);
+ igt_panfrost_free_bo(fd, submit->tiler_heap_bo);
+ igt_panfrost_free_bo(fd, submit->fbo);
+ free(submit->args);
+ free(submit);
+}
diff --git a/lib/igt_panfrost.h b/lib/igt_panfrost.h
new file mode 100644
index 000000000000..a33327e39736
--- /dev/null
+++ b/lib/igt_panfrost.h
@@ -0,0 +1,60 @@
+/*
+ * Copyright © 2016 Broadcom
+ * Copyright © 2019 Collabora, Ltd.
+ *
+ * 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.
+ */
+
+#ifndef IGT_PANFROST_H
+#define IGT_PANFROST_H
+
+#include "panfrost_drm.h"
+
+struct panfrost_bo {
+ int handle;
+ uint32_t offset;
+ uint32_t size;
+ void *map;
+};
+
+struct panfrost_submit {
+ struct drm_panfrost_submit *args;
+ struct panfrost_bo *submit_bo;
+ struct panfrost_bo *fb_bo;
+ struct panfrost_bo *scratchpad_bo;
+ struct panfrost_bo *tiler_scratch_bo;
+ struct panfrost_bo *tiler_heap_bo;
+ struct panfrost_bo *fbo;
+};
+
+struct panfrost_bo *igt_panfrost_gem_new(int fd, size_t size);
+void igt_panfrost_free_bo(int fd, struct panfrost_bo *bo);
+
+struct panfrost_submit *igt_panfrost_trivial_job(int fd, bool do_crash, int width, int height, uint32_t color);
+void igt_panfrost_free_job(int fd, struct panfrost_submit *submit);
+
+/* IOCTL wrappers */
+uint32_t igt_panfrost_get_bo_offset(int fd, uint32_t handle);
+uint32_t igt_panfrost_get_param(int fd, int param);
+void *igt_panfrost_mmap_bo(int fd, uint32_t handle, uint32_t size, unsigned prot);
+
+void igt_panfrost_bo_mmap(int fd, struct panfrost_bo *bo);
+
+#endif /* IGT_PANFROST_H */
diff --git a/lib/meson.build b/lib/meson.build
index 786e7a264f97..8073686867d4 100644
--- a/lib/meson.build
+++ b/lib/meson.build
@@ -52,6 +52,7 @@ lib_sources = [
'igt_dummyload.c',
'uwildmat/uwildmat.c',
'igt_kmod.c',
+ 'igt_panfrost.c',
'igt_v3d.c',
'igt_vc4.c',
'igt_psr.c',
diff --git a/lib/panfrost-job.h b/lib/panfrost-job.h
new file mode 100644
index 000000000000..85ef02d04e07
--- /dev/null
+++ b/lib/panfrost-job.h
@@ -0,0 +1,1516 @@
+/*
+ * © Copyright 2017-2018 Alyssa Rosenzweig
+ * © Copyright 2017-2018 Connor Abbott
+ * © Copyright 2017-2018 Lyude Paul
+ *
+ * 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.
+ *
+ */
+
+#ifndef __PANFROST_JOB_H__
+#define __PANFROST_JOB_H__
+
+#include <stdint.h>
+#include <panfrost-misc.h>
+
+#define MALI_SHORT_PTR_BITS (sizeof(uintptr_t)*8)
+
+#define MALI_FBD_HIERARCHY_WEIGHTS 8
+
+#define MALI_PAYLOAD_SIZE 256
+
+typedef u32 mali_jd_core_req;
+
+enum mali_job_type {
+ JOB_NOT_STARTED = 0,
+ JOB_TYPE_NULL = 1,
+ JOB_TYPE_SET_VALUE = 2,
+ JOB_TYPE_CACHE_FLUSH = 3,
+ JOB_TYPE_COMPUTE = 4,
+ JOB_TYPE_VERTEX = 5,
+ JOB_TYPE_GEOMETRY = 6,
+ JOB_TYPE_TILER = 7,
+ JOB_TYPE_FUSED = 8,
+ JOB_TYPE_FRAGMENT = 9,
+};
+
+enum mali_draw_mode {
+ MALI_DRAW_NONE = 0x0,
+ MALI_POINTS = 0x1,
+ MALI_LINES = 0x2,
+ MALI_LINE_STRIP = 0x4,
+ MALI_LINE_LOOP = 0x6,
+ MALI_TRIANGLES = 0x8,
+ MALI_TRIANGLE_STRIP = 0xA,
+ MALI_TRIANGLE_FAN = 0xC,
+ MALI_POLYGON = 0xD,
+ MALI_QUADS = 0xE,
+ MALI_QUAD_STRIP = 0xF,
+
+ /* All other modes invalid */
+};
+
+/* Applies to tiler_gl_enables */
+
+
+#define MALI_OCCLUSION_QUERY (1 << 3)
+#define MALI_OCCLUSION_PRECISE (1 << 4)
+
+#define MALI_FRONT_FACE(v) (v << 5)
+#define MALI_CCW (0)
+#define MALI_CW (1)
+
+#define MALI_CULL_FACE_FRONT (1 << 6)
+#define MALI_CULL_FACE_BACK (1 << 7)
+
+/* TODO: Might this actually be a finer bitfield? */
+#define MALI_DEPTH_STENCIL_ENABLE 0x6400
+
+#define DS_ENABLE(field) \
+ (field == MALI_DEPTH_STENCIL_ENABLE) \
+ ? "MALI_DEPTH_STENCIL_ENABLE" \
+ : (field == 0) ? "0" \
+ : "0 /* XXX: Unknown, check hexdump */"
+
+/* Used in stencil and depth tests */
+
+enum mali_func {
+ MALI_FUNC_NEVER = 0,
+ MALI_FUNC_LESS = 1,
+ MALI_FUNC_EQUAL = 2,
+ MALI_FUNC_LEQUAL = 3,
+ MALI_FUNC_GREATER = 4,
+ MALI_FUNC_NOTEQUAL = 5,
+ MALI_FUNC_GEQUAL = 6,
+ MALI_FUNC_ALWAYS = 7
+};
+
+/* Same OpenGL, but mixed up. Why? Because forget me, that's why! */
+
+enum mali_alt_func {
+ MALI_ALT_FUNC_NEVER = 0,
+ MALI_ALT_FUNC_GREATER = 1,
+ MALI_ALT_FUNC_EQUAL = 2,
+ MALI_ALT_FUNC_GEQUAL = 3,
+ MALI_ALT_FUNC_LESS = 4,
+ MALI_ALT_FUNC_NOTEQUAL = 5,
+ MALI_ALT_FUNC_LEQUAL = 6,
+ MALI_ALT_FUNC_ALWAYS = 7
+};
+
+/* Flags apply to unknown2_3? */
+
+#define MALI_HAS_MSAA (1 << 0)
+#define MALI_CAN_DISCARD (1 << 5)
+
+/* Applies on SFBD systems, specifying that programmable blending is in use */
+#define MALI_HAS_BLEND_SHADER (1 << 6)
+
+/* func is mali_func */
+#define MALI_DEPTH_FUNC(func) (func << 8)
+#define MALI_GET_DEPTH_FUNC(flags) ((flags >> 8) & 0x7)
+#define MALI_DEPTH_FUNC_MASK MALI_DEPTH_FUNC(0x7)
+
+#define MALI_DEPTH_TEST (1 << 11)
+
+/* Next flags to unknown2_4 */
+#define MALI_STENCIL_TEST (1 << 0)
+
+/* What?! */
+#define MALI_SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER (1 << 1)
+
+#define MALI_NO_DITHER (1 << 9)
+#define MALI_DEPTH_RANGE_A (1 << 12)
+#define MALI_DEPTH_RANGE_B (1 << 13)
+#define MALI_NO_MSAA (1 << 14)
+
+/* Stencil test state is all encoded in a single u32, just with a lot of
+ * enums... */
+
+enum mali_stencil_op {
+ MALI_STENCIL_KEEP = 0,
+ MALI_STENCIL_REPLACE = 1,
+ MALI_STENCIL_ZERO = 2,
+ MALI_STENCIL_INVERT = 3,
+ MALI_STENCIL_INCR_WRAP = 4,
+ MALI_STENCIL_DECR_WRAP = 5,
+ MALI_STENCIL_INCR = 6,
+ MALI_STENCIL_DECR = 7
+};
+
+struct mali_stencil_test {
+ unsigned ref : 8;
+ unsigned mask : 8;
+ enum mali_func func : 3;
+ enum mali_stencil_op sfail : 3;
+ enum mali_stencil_op dpfail : 3;
+ enum mali_stencil_op dppass : 3;
+ unsigned zero : 4;
+} __attribute__((packed));
+
+/* Blending is a mess, since anything fancy triggers a blend shader, and
+ * -those- are not understood whatsover yet */
+
+#define MALI_MASK_R (1 << 0)
+#define MALI_MASK_G (1 << 1)
+#define MALI_MASK_B (1 << 2)
+#define MALI_MASK_A (1 << 3)
+
+enum mali_nondominant_mode {
+ MALI_BLEND_NON_MIRROR = 0,
+ MALI_BLEND_NON_ZERO = 1
+};
+
+enum mali_dominant_blend {
+ MALI_BLEND_DOM_SOURCE = 0,
+ MALI_BLEND_DOM_DESTINATION = 1
+};
+
+enum mali_dominant_factor {
+ MALI_DOMINANT_UNK0 = 0,
+ MALI_DOMINANT_ZERO = 1,
+ MALI_DOMINANT_SRC_COLOR = 2,
+ MALI_DOMINANT_DST_COLOR = 3,
+ MALI_DOMINANT_UNK4 = 4,
+ MALI_DOMINANT_SRC_ALPHA = 5,
+ MALI_DOMINANT_DST_ALPHA = 6,
+ MALI_DOMINANT_CONSTANT = 7,
+};
+
+enum mali_blend_modifier {
+ MALI_BLEND_MOD_UNK0 = 0,
+ MALI_BLEND_MOD_NORMAL = 1,
+ MALI_BLEND_MOD_SOURCE_ONE = 2,
+ MALI_BLEND_MOD_DEST_ONE = 3,
+};
+
+struct mali_blend_mode {
+ enum mali_blend_modifier clip_modifier : 2;
+ unsigned unused_0 : 1;
+ unsigned negate_source : 1;
+
+ enum mali_dominant_blend dominant : 1;
+
+ enum mali_nondominant_mode nondominant_mode : 1;
+
+ unsigned unused_1 : 1;
+
+ unsigned negate_dest : 1;
+
+ enum mali_dominant_factor dominant_factor : 3;
+ unsigned complement_dominant : 1;
+} __attribute__((packed));
+
+struct mali_blend_equation {
+ /* Of type mali_blend_mode */
+ unsigned rgb_mode : 12;
+ unsigned alpha_mode : 12;
+
+ unsigned zero1 : 4;
+
+ /* Corresponds to MALI_MASK_* above and glColorMask arguments */
+
+ unsigned color_mask : 4;
+
+ /* Attached constant for CONSTANT_ALPHA, etc */
+
+#ifndef BIFROST
+ float constant;
+#endif
+} __attribute__((packed));
+
+/* Used with channel swizzling */
+enum mali_channel {
+ MALI_CHANNEL_RED = 0,
+ MALI_CHANNEL_GREEN = 1,
+ MALI_CHANNEL_BLUE = 2,
+ MALI_CHANNEL_ALPHA = 3,
+ MALI_CHANNEL_ZERO = 4,
+ MALI_CHANNEL_ONE = 5,
+ MALI_CHANNEL_RESERVED_0 = 6,
+ MALI_CHANNEL_RESERVED_1 = 7,
+};
+
+struct mali_channel_swizzle {
+ enum mali_channel r : 3;
+ enum mali_channel g : 3;
+ enum mali_channel b : 3;
+ enum mali_channel a : 3;
+} __attribute__((packed));
+
+/* Compressed per-pixel formats. Each of these formats expands to one to four
+ * floating-point or integer numbers, as defined by the OpenGL specification.
+ * There are various places in OpenGL where the user can specify a compressed
+ * format in memory, which all use the same 8-bit enum in the various
+ * descriptors, although different hardware units support different formats.
+ */
+
+/* The top 3 bits specify how the bits of each component are interpreted. */
+
+/* e.g. R11F_G11F_B10F */
+#define MALI_FORMAT_SPECIAL (2 << 5)
+
+/* signed normalized, e.g. RGBA8_SNORM */
+#define MALI_FORMAT_SNORM (3 << 5)
+
+/* e.g. RGBA8UI */
+#define MALI_FORMAT_UINT (4 << 5)
+
+/* e.g. RGBA8 and RGBA32F */
+#define MALI_FORMAT_UNORM (5 << 5)
+
+/* e.g. RGBA8I and RGBA16F */
+#define MALI_FORMAT_SINT (6 << 5)
+
+/* These formats seem to largely duplicate the others. They're used at least
+ * for Bifrost framebuffer output.
+ */
+#define MALI_FORMAT_SPECIAL2 (7 << 5)
+
+/* If the high 3 bits are 3 to 6 these two bits say how many components
+ * there are.
+ */
+#define MALI_NR_CHANNELS(n) ((n - 1) << 3)
+
+/* If the high 3 bits are 3 to 6, then the low 3 bits say how big each
+ * component is, except the special MALI_CHANNEL_FLOAT which overrides what the
+ * bits mean.
+ */
+
+#define MALI_CHANNEL_4 2
+
+#define MALI_CHANNEL_8 3
+
+#define MALI_CHANNEL_16 4
+
+#define MALI_CHANNEL_32 5
+
+/* For MALI_FORMAT_SINT it means a half-float (e.g. RG16F). For
+ * MALI_FORMAT_UNORM, it means a 32-bit float.
+ */
+#define MALI_CHANNEL_FLOAT 7
+
+enum mali_format {
+ MALI_RGB565 = MALI_FORMAT_SPECIAL | 0x0,
+ MALI_RGB5_A1_UNORM = MALI_FORMAT_SPECIAL | 0x2,
+ MALI_RGB10_A2_UNORM = MALI_FORMAT_SPECIAL | 0x3,
+ MALI_RGB10_A2_SNORM = MALI_FORMAT_SPECIAL | 0x5,
+ MALI_RGB10_A2UI = MALI_FORMAT_SPECIAL | 0x7,
+ MALI_RGB10_A2I = MALI_FORMAT_SPECIAL | 0x9,
+
+ /* YUV formats */
+ MALI_NV12 = MALI_FORMAT_SPECIAL | 0xc,
+
+ MALI_Z32_UNORM = MALI_FORMAT_SPECIAL | 0xD,
+ MALI_R32_FIXED = MALI_FORMAT_SPECIAL | 0x11,
+ MALI_RG32_FIXED = MALI_FORMAT_SPECIAL | 0x12,
+ MALI_RGB32_FIXED = MALI_FORMAT_SPECIAL | 0x13,
+ MALI_RGBA32_FIXED = MALI_FORMAT_SPECIAL | 0x14,
+ MALI_R11F_G11F_B10F = MALI_FORMAT_SPECIAL | 0x19,
+ /* Only used for varyings, to indicate the transformed gl_Position */
+ MALI_VARYING_POS = MALI_FORMAT_SPECIAL | 0x1e,
+ /* Only used for varyings, to indicate that the write should be
+ * discarded.
+ */
+ MALI_VARYING_DISCARD = MALI_FORMAT_SPECIAL | 0x1f,
+
+ MALI_R8_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(1) | MALI_CHANNEL_8,
+ MALI_R16_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(1) | MALI_CHANNEL_16,
+ MALI_R32_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(1) | MALI_CHANNEL_32,
+ MALI_RG8_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(2) | MALI_CHANNEL_8,
+ MALI_RG16_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(2) | MALI_CHANNEL_16,
+ MALI_RG32_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(2) | MALI_CHANNEL_32,
+ MALI_RGB8_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(3) | MALI_CHANNEL_8,
+ MALI_RGB16_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(3) | MALI_CHANNEL_16,
+ MALI_RGB32_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(3) | MALI_CHANNEL_32,
+ MALI_RGBA8_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(4) | MALI_CHANNEL_8,
+ MALI_RGBA16_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(4) | MALI_CHANNEL_16,
+ MALI_RGBA32_SNORM = MALI_FORMAT_SNORM | MALI_NR_CHANNELS(4) | MALI_CHANNEL_32,
+
+ MALI_R8UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(1) | MALI_CHANNEL_8,
+ MALI_R16UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(1) | MALI_CHANNEL_16,
+ MALI_R32UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(1) | MALI_CHANNEL_32,
+ MALI_RG8UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(2) | MALI_CHANNEL_8,
+ MALI_RG16UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(2) | MALI_CHANNEL_16,
+ MALI_RG32UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(2) | MALI_CHANNEL_32,
+ MALI_RGB8UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(3) | MALI_CHANNEL_8,
+ MALI_RGB16UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(3) | MALI_CHANNEL_16,
+ MALI_RGB32UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(3) | MALI_CHANNEL_32,
+ MALI_RGBA8UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(4) | MALI_CHANNEL_8,
+ MALI_RGBA16UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(4) | MALI_CHANNEL_16,
+ MALI_RGBA32UI = MALI_FORMAT_UINT | MALI_NR_CHANNELS(4) | MALI_CHANNEL_32,
+
+ MALI_R8_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(1) | MALI_CHANNEL_8,
+ MALI_R16_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(1) | MALI_CHANNEL_16,
+ MALI_R32_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(1) | MALI_CHANNEL_32,
+ MALI_R32F = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(1) | MALI_CHANNEL_FLOAT,
+ MALI_RG8_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(2) | MALI_CHANNEL_8,
+ MALI_RG16_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(2) | MALI_CHANNEL_16,
+ MALI_RG32_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(2) | MALI_CHANNEL_32,
+ MALI_RG32F = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(2) | MALI_CHANNEL_FLOAT,
+ MALI_RGB8_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(3) | MALI_CHANNEL_8,
+ MALI_RGB16_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(3) | MALI_CHANNEL_16,
+ MALI_RGB32_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(3) | MALI_CHANNEL_32,
+ MALI_RGB32F = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(3) | MALI_CHANNEL_FLOAT,
+ MALI_RGBA4_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(4) | MALI_CHANNEL_4,
+ MALI_RGBA8_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(4) | MALI_CHANNEL_8,
+ MALI_RGBA16_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(4) | MALI_CHANNEL_16,
+ MALI_RGBA32_UNORM = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(4) | MALI_CHANNEL_32,
+ MALI_RGBA32F = MALI_FORMAT_UNORM | MALI_NR_CHANNELS(4) | MALI_CHANNEL_FLOAT,
+
+ MALI_R8I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(1) | MALI_CHANNEL_8,
+ MALI_R16I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(1) | MALI_CHANNEL_16,
+ MALI_R32I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(1) | MALI_CHANNEL_32,
+ MALI_R16F = MALI_FORMAT_SINT | MALI_NR_CHANNELS(1) | MALI_CHANNEL_FLOAT,
+ MALI_RG8I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(2) | MALI_CHANNEL_8,
+ MALI_RG16I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(2) | MALI_CHANNEL_16,
+ MALI_RG32I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(2) | MALI_CHANNEL_32,
+ MALI_RG16F = MALI_FORMAT_SINT | MALI_NR_CHANNELS(2) | MALI_CHANNEL_FLOAT,
+ MALI_RGB8I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(3) | MALI_CHANNEL_8,
+ MALI_RGB16I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(3) | MALI_CHANNEL_16,
+ MALI_RGB32I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(3) | MALI_CHANNEL_32,
+ MALI_RGB16F = MALI_FORMAT_SINT | MALI_NR_CHANNELS(3) | MALI_CHANNEL_FLOAT,
+ MALI_RGBA8I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(4) | MALI_CHANNEL_8,
+ MALI_RGBA16I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(4) | MALI_CHANNEL_16,
+ MALI_RGBA32I = MALI_FORMAT_SINT | MALI_NR_CHANNELS(4) | MALI_CHANNEL_32,
+ MALI_RGBA16F = MALI_FORMAT_SINT | MALI_NR_CHANNELS(4) | MALI_CHANNEL_FLOAT,
+
+ MALI_RGBA4 = MALI_FORMAT_SPECIAL2 | 0x8,
+ MALI_RGBA8_2 = MALI_FORMAT_SPECIAL2 | 0xd,
+ MALI_RGB10_A2_2 = MALI_FORMAT_SPECIAL2 | 0xe,
+};
+
+
+/* Alpha coverage is encoded as 4-bits (from a clampf), with inversion
+ * literally performing a bitwise invert. This function produces slightly wrong
+ * results and I'm not sure why; some rounding issue I suppose... */
+
+#define MALI_ALPHA_COVERAGE(clampf) ((uint16_t) (int) (clampf * 15.0f))
+#define MALI_GET_ALPHA_COVERAGE(nibble) ((float) nibble / 15.0f)
+
+/* Applies to unknown1 */
+#define MALI_NO_ALPHA_TO_COVERAGE (1 << 10)
+
+struct mali_blend_meta {
+#ifndef BIFROST
+ /* Base value of 0x200.
+ * OR with 0x1 for blending (anything other than REPLACE).
+ * OR with 0x2 for programmable blending
+ */
+
+ u64 unk1;
+
+ /* For programmable blending, these turn into the blend_shader address */
+ struct mali_blend_equation blend_equation_1;
+
+ u64 zero2;
+ struct mali_blend_equation blend_equation_2;
+#else
+ u32 unk1; // = 0x200
+ struct mali_blend_equation blend_equation;
+ /*
+ * - 0x19 normally
+ * - 0x3 when this slot is unused (everything else is 0 except the index)
+ * - 0x11 when this is the fourth slot (and it's used)
++ * - 0 when there is a blend shader
+ */
+ u16 unk2;
+ /* increments from 0 to 3 */
+ u16 index;
+
+ union {
+ struct {
+ /* So far, I've only seen:
+ * - R001 for 1-component formats
+ * - RG01 for 2-component formats
+ * - RGB1 for 3-component formats
+ * - RGBA for 4-component formats
+ */
+ u32 swizzle : 12;
+ enum mali_format format : 8;
+
+ /* Type of the shader output variable. Note, this can
+ * be different from the format.
+ *
+ * 0: f16 (mediump float)
+ * 1: f32 (highp float)
+ * 2: i32 (highp int)
+ * 3: u32 (highp uint)
+ * 4: i16 (mediump int)
+ * 5: u16 (mediump uint)
+ */
+ u32 shader_type : 3;
+ u32 zero : 9;
+ };
+
+ /* Only the low 32 bits of the blend shader are stored, the
+ * high 32 bits are implicitly the same as the original shader.
+ * According to the kernel driver, the program counter for
+ * shaders is actually only 24 bits, so shaders cannot cross
+ * the 2^24-byte boundary, and neither can the blend shader.
+ * The blob handles this by allocating a 2^24 byte pool for
+ * shaders, and making sure that any blend shaders are stored
+ * in the same pool as the original shader. The kernel will
+ * make sure this allocation is aligned to 2^24 bytes.
+ */
+ u32 blend_shader;
+ };
+#endif
+} __attribute__((packed));
+
+struct mali_shader_meta {
+ mali_ptr shader;
+ u16 texture_count;
+ u16 sampler_count;
+ u16 attribute_count;
+ u16 varying_count;
+
+ union {
+ struct {
+ u32 uniform_buffer_count : 4;
+ u32 unk1 : 28; // = 0x800000 for vertex, 0x958020 for tiler
+ } bifrost1;
+ struct {
+ /* 0x200 except MALI_NO_ALPHA_TO_COVERAGE. Mysterious 1
+ * other times. Who knows really? */
+ u16 unknown1;
+
+ /* Whole number of uniform registers used, times two;
+ * whole number of work registers used (no scale).
+ */
+ unsigned work_count : 5;
+ unsigned uniform_count : 5;
+ unsigned unknown2 : 6;
+ } midgard1;
+ };
+
+ /* On bifrost: Exactly the same as glPolygonOffset() for both.
+ * On midgard: Depth factor is exactly as passed to glPolygonOffset.
+ * Depth units is equal to the value passed to glDeptOhffset + 1.0f
+ * (use MALI_NEGATIVE)
+ */
+ float depth_units;
+ float depth_factor;
+
+ u32 unknown2_2;
+
+ u16 alpha_coverage;
+ u16 unknown2_3;
+
+ u8 stencil_mask_front;
+ u8 stencil_mask_back;
+ u16 unknown2_4;
+
+ struct mali_stencil_test stencil_front;
+ struct mali_stencil_test stencil_back;
+
+ union {
+ struct {
+ u32 unk3 : 7;
+ /* On Bifrost, some system values are preloaded in
+ * registers R55-R62 by the thread dispatcher prior to
+ * the start of shader execution. This is a bitfield
+ * with one entry for each register saying which
+ * registers need to be preloaded. Right now, the known
+ * values are:
+ *
+ * Vertex/compute:
+ * - R55 : gl_LocalInvocationID.xy
+ * - R56 : gl_LocalInvocationID.z + unknown in high 16 bits
+ * - R57 : gl_WorkGroupID.x
+ * - R58 : gl_WorkGroupID.y
+ * - R59 : gl_WorkGroupID.z
+ * - R60 : gl_GlobalInvocationID.x
+ * - R61 : gl_GlobalInvocationID.y/gl_VertexID (without base)
+ * - R62 : gl_GlobalInvocationID.z/gl_InstanceID (without base)
+ *
+ * Fragment:
+ * - R55 : unknown, never seen (but the bit for this is
+ * always set?)
+ * - R56 : unknown (bit always unset)
+ * - R57 : gl_PrimitiveID
+ * - R58 : gl_FrontFacing in low bit, potentially other stuff
+ * - R59 : u16 fragment coordinates (used to compute
+ * gl_FragCoord.xy, together with sample positions)
+ * - R60 : gl_SampleMask (used in epilog, so pretty
+ * much always used, but the bit is always 0 -- is
+ * this just always pushed?)
+ * - R61 : gl_SampleMaskIn and gl_SampleID, used by
+ * varying interpolation.
+ * - R62 : unknown (bit always unset).
+ */
+ u32 preload_regs : 8;
+ /* In units of 8 bytes or 64 bits, since the
+ * uniform/const port loads 64 bits at a time.
+ */
+ u32 uniform_count : 7;
+ u32 unk4 : 10; // = 2
+ } bifrost2;
+ struct {
+ u32 unknown2_7;
+ } midgard2;
+ };
+
+ /* zero on bifrost */
+ u32 unknown2_8;
+
+ /* Blending information for the older non-MRT Midgard HW. Check for
+ * MALI_HAS_BLEND_SHADER to decide how to interpret.
+ */
+
+ union {
+ mali_ptr blend_shader;
+ struct mali_blend_equation blend_equation;
+ };
+
+ /* There can be up to 4 blend_meta's. None of them are required for
+ * vertex shaders or the non-MRT case for Midgard (so the blob doesn't
+ * allocate any space).
+ */
+ struct mali_blend_meta blend_meta[];
+
+} __attribute__((packed));
+
+/* This only concerns hardware jobs */
+
+/* Possible values for job_descriptor_size */
+
+#define MALI_JOB_32 0
+#define MALI_JOB_64 1
+
+struct mali_job_descriptor_header {
+ u32 exception_status;
+ u32 first_incomplete_task;
+ u64 fault_pointer;
+ u8 job_descriptor_size : 1;
+ enum mali_job_type job_type : 7;
+ u8 job_barrier : 1;
+ u8 unknown_flags : 7;
+ u16 job_index;
+ u16 job_dependency_index_1;
+ u16 job_dependency_index_2;
+
+ union {
+ u64 next_job_64;
+ u32 next_job_32;
+ };
+} __attribute__((packed));
+
+struct mali_payload_set_value {
+ u64 out;
+ u64 unknown;
+} __attribute__((packed));
+
+/* Special attributes have a fixed index */
+#define MALI_SPECIAL_ATTRIBUTE_BASE 16
+#define MALI_VERTEX_ID (MALI_SPECIAL_ATTRIBUTE_BASE + 0)
+#define MALI_INSTANCE_ID (MALI_SPECIAL_ATTRIBUTE_BASE + 1)
+
+/*
+ * Mali Attributes
+ *
+ * This structure lets the attribute unit compute the address of an attribute
+ * given the vertex and instance ID. Unfortunately, the way this works is
+ * rather complicated when instancing is enabled.
+ *
+ * To explain this, first we need to explain how compute and vertex threads are
+ * dispatched. This is a guess (although a pretty firm guess!) since the
+ * details are mostly hidden from the driver, except for attribute instancing.
+ * When a quad is dispatched, it receives a single, linear index. However, we
+ * need to translate that index into a (vertex id, instance id) pair, or a
+ * (local id x, local id y, local id z) triple for compute shaders (although
+ * vertex shaders and compute shaders are handled almost identically).
+ * Focusing on vertex shaders, one option would be to do:
+ *
+ * vertex_id = linear_id % num_vertices
+ * instance_id = linear_id / num_vertices
+ *
+ * but this involves a costly division and modulus by an arbitrary number.
+ * Instead, we could pad num_vertices. We dispatch padded_num_vertices *
+ * num_instances threads instead of num_vertices * num_instances, which results
+ * in some "extra" threads with vertex_id >= num_vertices, which we have to
+ * discard. The more we pad num_vertices, the more "wasted" threads we
+ * dispatch, but the division is potentially easier.
+ *
+ * One straightforward choice is to pad num_vertices to the next power of two,
+ * which means that the division and modulus are just simple bit shifts and
+ * masking. But the actual algorithm is a bit more complicated. The thread
+ * dispatcher has special support for dividing by 3, 5, 7, and 9, in addition
+ * to dividing by a power of two. This is possibly using the technique
+ * described in patent US20170010862A1. As a result, padded_num_vertices can be
+ * 1, 3, 5, 7, or 9 times a power of two. This results in less wasted threads,
+ * since we need less padding.
+ *
+ * padded_num_vertices is picked by the hardware. The driver just specifies the
+ * actual number of vertices. At least for Mali G71, the first few cases are
+ * given by:
+ *
+ * num_vertices | padded_num_vertices
+ * 3 | 4
+ * 4-7 | 8
+ * 8-11 | 12 (3 * 4)
+ * 12-15 | 16
+ * 16-19 | 20 (5 * 4)
+ *
+ * Note that padded_num_vertices is a multiple of four (presumably because
+ * threads are dispatched in groups of 4). Also, padded_num_vertices is always
+ * at least one more than num_vertices, which seems like a quirk of the
+ * hardware. For larger num_vertices, the hardware uses the following
+ * algorithm: using the binary representation of num_vertices, we look at the
+ * most significant set bit as well as the following 3 bits. Let n be the
+ * number of bits after those 4 bits. Then we set padded_num_vertices according
+ * to the following table:
+ *
+ * high bits | padded_num_vertices
+ * 1000 | 9 * 2^n
+ * 1001 | 5 * 2^(n+1)
+ * 101x | 3 * 2^(n+2)
+ * 110x | 7 * 2^(n+1)
+ * 111x | 2^(n+4)
+ *
+ * For example, if num_vertices = 70 is passed to glDraw(), its binary
+ * representation is 1000110, so n = 3 and the high bits are 1000, and
+ * therefore padded_num_vertices = 9 * 2^3 = 72.
+ *
+ * The attribute unit works in terms of the original linear_id. if
+ * num_instances = 1, then they are the same, and everything is simple.
+ * However, with instancing things get more complicated. There are four
+ * possible modes, two of them we can group together:
+ *
+ * 1. Use the linear_id directly. Only used when there is no instancing.
+ *
+ * 2. Use the linear_id modulo a constant. This is used for per-vertex
+ * attributes with instancing enabled by making the constant equal
+ * padded_num_vertices. Because the modulus is always padded_num_vertices, this
+ * mode only supports a modulus that is a power of 2 times 1, 3, 5, 7, or 9.
+ * The shift field specifies the power of two, while the extra_flags field
+ * specifies the odd number. If shift = n and extra_flags = m, then the modulus
+ * is (2m + 1) * 2^n. As an example, if num_vertices = 70, then as computed
+ * above, padded_num_vertices = 9 * 2^3, so we should set extra_flags = 4 and
+ * shift = 3. Note that we must exactly follow the hardware algorithm used to
+ * get padded_num_vertices in order to correctly implement per-vertex
+ * attributes.
+ *
+ * 3. Divide the linear_id by a constant. In order to correctly implement
+ * instance divisors, we have to divide linear_id by padded_num_vertices times
+ * to user-specified divisor. So first we compute padded_num_vertices, again
+ * following the exact same algorithm that the hardware uses, then multiply it
+ * by the GL-level divisor to get the hardware-level divisor. This case is
+ * further divided into two more cases. If the hardware-level divisor is a
+ * power of two, then we just need to shift. The shift amount is specified by
+ * the shift field, so that the hardware-level divisor is just 2^shift.
+ *
+ * If it isn't a power of two, then we have to divide by an arbitrary integer.
+ * For that, we use the well-known technique of multiplying by an approximation
+ * of the inverse. The driver must compute the magic multiplier and shift
+ * amount, and then the hardware does the multiplication and shift. The
+ * hardware and driver also use the "round-down" optimization as described in
+ * http://ridiculousfish.com/files/faster_unsigned_division_by_constants.pdf.
+ * The hardware further assumes the multiplier is between 2^31 and 2^32, so the
+ * high bit is implicitly set to 1 even though it is set to 0 by the driver --
+ * presumably this simplifies the hardware multiplier a little. The hardware
+ * first multiplies linear_id by the multiplier and takes the high 32 bits,
+ * then applies the round-down correction if extra_flags = 1, then finally
+ * shifts right by the shift field.
+ *
+ * There are some differences between ridiculousfish's algorithm and the Mali
+ * hardware algorithm, which means that the reference code from ridiculousfish
+ * doesn't always produce the right constants. Mali does not use the pre-shift
+ * optimization, since that would make a hardware implementation slower (it
+ * would have to always do the pre-shift, multiply, and post-shift operations).
+ * It also forces the multplier to be at least 2^31, which means that the
+ * exponent is entirely fixed, so there is no trial-and-error. Altogether,
+ * given the divisor d, the algorithm the driver must follow is:
+ *
+ * 1. Set shift = floor(log2(d)).
+ * 2. Compute m = ceil(2^(shift + 32) / d) and e = 2^(shift + 32) % d.
+ * 3. If e <= 2^shift, then we need to use the round-down algorithm. Set
+ * magic_divisor = m - 1 and extra_flags = 1.
+ * 4. Otherwise, set magic_divisor = m and extra_flags = 0.
+ */
+
+enum mali_attr_mode {
+ MALI_ATTR_UNUSED = 0,
+ MALI_ATTR_LINEAR = 1,
+ MALI_ATTR_POT_DIVIDE = 2,
+ MALI_ATTR_MODULO = 3,
+ MALI_ATTR_NPOT_DIVIDE = 4,
+};
+
+union mali_attr {
+ /* This is used for actual attributes. */
+ struct {
+ /* The bottom 3 bits are the mode */
+ mali_ptr elements : 64 - 8;
+ u32 shift : 5;
+ u32 extra_flags : 3;
+ u32 stride;
+ u32 size;
+ };
+ /* The entry after an NPOT_DIVIDE entry has this format. It stores
+ * extra information that wouldn't fit in a normal entry.
+ */
+ struct {
+ u32 unk; /* = 0x20 */
+ u32 magic_divisor;
+ u32 zero;
+ /* This is the original, GL-level divisor. */
+ u32 divisor;
+ };
+} __attribute__((packed));
+
+struct mali_attr_meta {
+ /* Vertex buffer index */
+ u8 index;
+
+ unsigned unknown1 : 2;
+ unsigned swizzle : 12;
+ enum mali_format format : 8;
+
+ /* Always observed to be zero at the moment */
+ unsigned unknown3 : 2;
+
+ /* When packing multiple attributes in a buffer, offset addresses by this value */
+ uint32_t src_offset;
+} __attribute__((packed));
+
+enum mali_fbd_type {
+ MALI_SFBD = 0,
+ MALI_MFBD = 1,
+};
+
+#define FBD_TYPE (1)
+#define FBD_MASK (~0x3f)
+
+struct mali_uniform_buffer_meta {
+ /* This is actually the size minus 1 (MALI_POSITIVE), in units of 16
+ * bytes. This gives a maximum of 2^14 bytes, which just so happens to
+ * be the GL minimum-maximum for GL_MAX_UNIFORM_BLOCK_SIZE.
+ */
+ u64 size : 10;
+
+ /* This is missing the bottom 2 bits and top 8 bits. The top 8 bits
+ * should be 0 for userspace pointers, according to
+ * https://lwn.net/Articles/718895/. By reusing these bits, we can make
+ * each entry in the table only 64 bits.
+ */
+ mali_ptr ptr : 64 - 10;
+};
+
+/* On Bifrost, these fields are the same between the vertex and tiler payloads.
+ * They also seem to be the same between Bifrost and Midgard. They're shared in
+ * fused payloads.
+ */
+
+/* Applies to unknown_draw */
+
+#define MALI_DRAW_INDEXED_UINT8 (0x10)
+#define MALI_DRAW_INDEXED_UINT16 (0x20)
+#define MALI_DRAW_INDEXED_UINT32 (0x30)
+#define MALI_DRAW_VARYING_SIZE (0x100)
+#define MALI_DRAW_PRIMITIVE_RESTART_FIXED_INDEX (0x10000)
+
+struct mali_vertex_tiler_prefix {
+ /* This is a dynamic bitfield containing the following things in this order:
+ *
+ * - gl_WorkGroupSize.x
+ * - gl_WorkGroupSize.y
+ * - gl_WorkGroupSize.z
+ * - gl_NumWorkGroups.x
+ * - gl_NumWorkGroups.y
+ * - gl_NumWorkGroups.z
+ *
+ * The number of bits allocated for each number is based on the *_shift
+ * fields below. For example, workgroups_y_shift gives the bit that
+ * gl_NumWorkGroups.y starts at, and workgroups_z_shift gives the bit
+ * that gl_NumWorkGroups.z starts at (and therefore one after the bit
+ * that gl_NumWorkGroups.y ends at). The actual value for each gl_*
+ * value is one more than the stored value, since if any of the values
+ * are zero, then there would be no invocations (and hence no job). If
+ * there were 0 bits allocated to a given field, then it must be zero,
+ * and hence the real value is one.
+ *
+ * Vertex jobs reuse the same job dispatch mechanism as compute jobs,
+ * effectively doing glDispatchCompute(1, vertex_count, instance_count)
+ * where vertex count is the number of vertices.
+ */
+ u32 invocation_count;
+
+ u32 size_y_shift : 5;
+ u32 size_z_shift : 5;
+ u32 workgroups_x_shift : 6;
+ u32 workgroups_y_shift : 6;
+ u32 workgroups_z_shift : 6;
+ /* This is max(workgroups_x_shift, 2) in all the cases I've seen. */
+ u32 workgroups_x_shift_2 : 4;
+
+ u32 draw_mode : 4;
+ u32 unknown_draw : 22;
+
+ /* This is the the same as workgroups_x_shift_2 in compute shaders, but
+ * always 5 for vertex jobs and 6 for tiler jobs. I suspect this has
+ * something to do with how many quads get put in the same execution
+ * engine, which is a balance (you don't want to starve the engine, but
+ * you also want to distribute work evenly).
+ */
+ u32 workgroups_x_shift_3 : 6;
+
+
+ /* Negative of draw_start for TILER jobs from what I've seen */
+ int32_t negative_start;
+ u32 zero1;
+
+ /* Like many other strictly nonzero quantities, index_count is
+ * subtracted by one. For an indexed cube, this is equal to 35 = 6
+ * faces * 2 triangles/per face * 3 vertices/per triangle - 1. That is,
+ * for an indexed draw, index_count is the number of actual vertices
+ * rendered whereas invocation_count is the number of unique vertices
+ * rendered (the number of times the vertex shader must be invoked).
+ * For non-indexed draws, this is just equal to invocation_count. */
+
+ u32 index_count;
+
+ /* No hidden structure; literally just a pointer to an array of uint
+ * indices (width depends on flags). Thanks, guys, for not making my
+ * life insane for once! NULL for non-indexed draws. */
+
+ uintptr_t indices;
+} __attribute__((packed));
+
+/* Point size / line width can either be specified as a 32-bit float (for
+ * constant size) or as a [machine word size]-bit GPU pointer (for varying size). If a pointer
+ * is selected, by setting the appropriate MALI_DRAW_VARYING_SIZE bit in the tiler
+ * payload, the contents of varying_pointer will be intepreted as an array of
+ * fp16 sizes, one for each vertex. gl_PointSize is therefore implemented by
+ * creating a special MALI_R16F varying writing to varying_pointer. */
+
+union midgard_primitive_size {
+ float constant;
+ uintptr_t pointer;
+};
+
+struct bifrost_vertex_only {
+ u32 unk2; /* =0x2 */
+
+ u32 zero0;
+
+ u64 zero1;
+} __attribute__((packed));
+
+struct bifrost_tiler_heap_meta {
+ u32 zero;
+ u32 heap_size;
+ /* note: these are just guesses! */
+ mali_ptr tiler_heap_start;
+ mali_ptr tiler_heap_free;
+ mali_ptr tiler_heap_end;
+
+ /* hierarchy weights? but they're still 0 after the job has run... */
+ u32 zeros[12];
+} __attribute__((packed));
+
+struct bifrost_tiler_meta {
+ u64 zero0;
+ u32 unk; // = 0xf0
+ u16 width;
+ u16 height;
+ u64 zero1;
+ mali_ptr tiler_heap_meta;
+ /* TODO what is this used for? */
+ u64 zeros[20];
+} __attribute__((packed));
+
+struct bifrost_tiler_only {
+ /* 0x20 */
+ union midgard_primitive_size primitive_size;
+
+ mali_ptr tiler_meta;
+
+ u64 zero1, zero2, zero3, zero4, zero5, zero6;
+
+ u32 gl_enables;
+ u32 zero7;
+ u64 zero8;
+} __attribute__((packed));
+
+struct bifrost_scratchpad {
+ u32 zero;
+ u32 flags; // = 0x1f
+ /* This is a pointer to a CPU-inaccessible buffer, 16 pages, allocated
+ * during startup. It seems to serve the same purpose as the
+ * gpu_scratchpad in the SFBD for Midgard, although it's slightly
+ * larger.
+ */
+ mali_ptr gpu_scratchpad;
+} __attribute__((packed));
+
+struct mali_vertex_tiler_postfix {
+ /* Zero for vertex jobs. Pointer to the position (gl_Position) varying
+ * output from the vertex shader for tiler jobs.
+ */
+
+ uintptr_t position_varying;
+
+ /* An array of mali_uniform_buffer_meta's. The size is given by the
+ * shader_meta.
+ */
+ uintptr_t uniform_buffers;
+
+ /* This is a pointer to an array of pointers to the texture
+ * descriptors, number of pointers bounded by number of textures. The
+ * indirection is needed to accomodate varying numbers and sizes of
+ * texture descriptors */
+ uintptr_t texture_trampoline;
+
+ /* For OpenGL, from what I've seen, this is intimately connected to
+ * texture_meta. cwabbott says this is not the case under Vulkan, hence
+ * why this field is seperate (Midgard is Vulkan capable). Pointer to
+ * array of sampler descriptors (which are uniform in size) */
+ uintptr_t sampler_descriptor;
+
+ uintptr_t uniforms;
+ u8 flags : 4;
+ uintptr_t _shader_upper : MALI_SHORT_PTR_BITS - 4; /* struct shader_meta */
+ uintptr_t attributes; /* struct attribute_buffer[] */
+ uintptr_t attribute_meta; /* attribute_meta[] */
+ uintptr_t varyings; /* struct attr */
+ uintptr_t varying_meta; /* pointer */
+ uintptr_t viewport;
+ uintptr_t occlusion_counter; /* A single bit as far as I can tell */
+
+ /* Note: on Bifrost, this isn't actually the FBD. It points to
+ * bifrost_scratchpad instead. However, it does point to the same thing
+ * in vertex and tiler jobs.
+ */
+ mali_ptr framebuffer;
+
+#ifdef __LP64__
+#ifdef BIFROST
+ /* most likely padding to make this a multiple of 64 bytes */
+ u64 zero7;
+#endif
+#endif
+} __attribute__((packed));
+
+struct midgard_payload_vertex_tiler {
+#ifndef __LP64__
+ union midgard_primitive_size primitive_size;
+#endif
+
+ struct mali_vertex_tiler_prefix prefix;
+
+#ifndef __LP64__
+ u32 zero3;
+#endif
+
+ u32 gl_enables; // 0x5
+
+ /* Offset for first vertex in buffer */
+ u32 draw_start;
+
+ uintptr_t zero5;
+
+ struct mali_vertex_tiler_postfix postfix;
+
+#ifdef __LP64__
+ union midgard_primitive_size primitive_size;
+#endif
+} __attribute__((packed));
+
+struct bifrost_payload_vertex {
+ struct mali_vertex_tiler_prefix prefix;
+ struct bifrost_vertex_only vertex;
+ struct mali_vertex_tiler_postfix postfix;
+} __attribute__((packed));
+
+struct bifrost_payload_tiler {
+ struct mali_vertex_tiler_prefix prefix;
+ struct bifrost_tiler_only tiler;
+ struct mali_vertex_tiler_postfix postfix;
+} __attribute__((packed));
+
+struct bifrost_payload_fused {
+ struct mali_vertex_tiler_prefix prefix;
+ struct bifrost_tiler_only tiler;
+ struct mali_vertex_tiler_postfix tiler_postfix;
+ struct bifrost_vertex_only vertex;
+ struct mali_vertex_tiler_postfix vertex_postfix;
+} __attribute__((packed));
+
+/* Pointed to from texture_trampoline, mostly unknown still, haven't
+ * managed to replay successfully */
+
+/* Purposeful off-by-one in width, height fields. For example, a (64, 64)
+ * texture is stored as (63, 63) in these fields. This adjusts for that.
+ * There's an identical pattern in the framebuffer descriptor. Even vertex
+ * count fields work this way, hence the generic name -- integral fields that
+ * are strictly positive generally need this adjustment. */
+
+#define MALI_POSITIVE(dim) (dim - 1)
+
+/* Opposite of MALI_POSITIVE, found in the depth_units field */
+
+#define MALI_NEGATIVE(dim) (dim + 1)
+
+/* Used with wrapping. Incomplete (this is a 4-bit field...) */
+
+enum mali_wrap_mode {
+ MALI_WRAP_REPEAT = 0x8,
+ MALI_WRAP_CLAMP_TO_EDGE = 0x9,
+ MALI_WRAP_CLAMP_TO_BORDER = 0xB,
+ MALI_WRAP_MIRRORED_REPEAT = 0xC
+};
+
+/* 8192x8192 */
+#define MAX_MIP_LEVELS (13)
+
+/* Cubemap bloats everything up */
+#define MAX_FACES (6)
+
+/* Corresponds to the type passed to glTexImage2D and so forth */
+
+struct mali_texture_format {
+ unsigned swizzle : 12;
+ enum mali_format format : 8;
+
+ unsigned usage1 : 3;
+ unsigned is_not_cubemap : 1;
+ unsigned usage2 : 8;
+} __attribute__((packed));
+
+struct mali_texture_descriptor {
+ uint16_t width;
+ uint16_t height;
+ uint16_t depth;
+
+ uint16_t unknown1;
+
+ struct mali_texture_format format;
+
+ uint16_t unknown3;
+
+ /* One for non-mipmapped, zero for mipmapped */
+ uint8_t unknown3A;
+
+ /* Zero for non-mipmapped, (number of levels - 1) for mipmapped */
+ uint8_t nr_mipmap_levels;
+
+ /* Swizzling is a single 32-bit word, broken up here for convenience.
+ * Here, swizzling refers to the ES 3.0 texture parameters for channel
+ * level swizzling, not the internal pixel-level swizzling which is
+ * below OpenGL's reach */
+
+ unsigned swizzle : 12;
+ unsigned swizzle_zero : 20;
+
+ uint32_t unknown5;
+ uint32_t unknown6;
+ uint32_t unknown7;
+
+ mali_ptr swizzled_bitmaps[MAX_MIP_LEVELS * MAX_FACES];
+} __attribute__((packed));
+
+/* Used as part of filter_mode */
+
+#define MALI_LINEAR 0
+#define MALI_NEAREST 1
+#define MALI_MIP_LINEAR (0x18)
+
+/* Used to construct low bits of filter_mode */
+
+#define MALI_TEX_MAG(mode) (((mode) & 1) << 0)
+#define MALI_TEX_MIN(mode) (((mode) & 1) << 1)
+
+#define MALI_TEX_MAG_MASK (1)
+#define MALI_TEX_MIN_MASK (2)
+
+#define MALI_FILTER_NAME(filter) (filter ? "MALI_NEAREST" : "MALI_LINEAR")
+
+/* Used for lod encoding. Thanks @urjaman for pointing out these routines can
+ * be cleaned up a lot. */
+
+#define DECODE_FIXED_16(x) ((float) (x / 256.0))
+
+static inline uint16_t
+FIXED_16(float x)
+{
+ /* Clamp inputs, accounting for float error */
+ float max_lod = (32.0 - (1.0 / 512.0));
+
+ x = ((x > max_lod) ? max_lod : ((x < 0.0) ? 0.0 : x));
+
+ return (int) (x * 256.0);
+}
+
+struct mali_sampler_descriptor {
+ uint32_t filter_mode;
+
+ /* Fixed point. Upper 8-bits is before the decimal point, although it
+ * caps [0-31]. Lower 8-bits is after the decimal point: int(round(x *
+ * 256)) */
+
+ uint16_t min_lod;
+ uint16_t max_lod;
+
+ /* All one word in reality, but packed a bit */
+
+ enum mali_wrap_mode wrap_s : 4;
+ enum mali_wrap_mode wrap_t : 4;
+ enum mali_wrap_mode wrap_r : 4;
+ enum mali_alt_func compare_func : 3;
+
+ /* A single set bit of unknown, ha! */
+ unsigned unknown2 : 1;
+
+ unsigned zero : 16;
+
+ uint32_t zero2;
+ float border_color[4];
+} __attribute__((packed));
+
+/* TODO: What are the floats? Apparently always { -inf, -inf, inf, inf },
+ * unless the scissor test is enabled.
+ *
+ * viewport0/viewport1 form the arguments to glViewport. viewport1 is modified
+ * by MALI_POSITIVE; viewport0 is as-is.
+ */
+
+struct mali_viewport {
+ /* XY clipping planes */
+ float clip_minx;
+ float clip_miny;
+ float clip_maxx;
+ float clip_maxy;
+
+ /* Depth clipping planes */
+ float clip_minz;
+ float clip_maxz;
+
+ u16 viewport0[2];
+ u16 viewport1[2];
+} __attribute__((packed));
+
+/* TODO: Varying meta is symmetrical with attr_meta, but there is some
+ * weirdness associated. Figure it out. */
+
+struct mali_unknown6 {
+ u64 unknown0;
+ u64 unknown1;
+};
+
+/* From presentations, 16x16 tiles externally. Use shift for fast computation
+ * of tile numbers. */
+
+#define MALI_TILE_SHIFT 4
+#define MALI_TILE_LENGTH (1 << MALI_TILE_SHIFT)
+
+/* Tile coordinates are stored as a compact u32, as only 12 bits are needed to
+ * each component. Notice that this provides a theoretical upper bound of (1 <<
+ * 12) = 4096 tiles in each direction, addressing a maximum framebuffer of size
+ * 65536x65536. Multiplying that together, times another four given that Mali
+ * framebuffers are 32-bit ARGB8888, means that this upper bound would take 16
+ * gigabytes of RAM just to store the uncompressed framebuffer itself, let
+ * alone rendering in real-time to such a buffer.
+ *
+ * Nice job, guys.*/
+
+/* From mali_kbase_10969_workaround.c */
+#define MALI_X_COORD_MASK 0x00000FFF
+#define MALI_Y_COORD_MASK 0x0FFF0000
+
+/* Extract parts of a tile coordinate */
+
+#define MALI_TILE_COORD_X(coord) ((coord) & MALI_X_COORD_MASK)
+#define MALI_TILE_COORD_Y(coord) (((coord) & MALI_Y_COORD_MASK) >> 16)
+#define MALI_TILE_COORD_FLAGS(coord) ((coord) & ~(MALI_X_COORD_MASK | MALI_Y_COORD_MASK))
+
+/* No known flags yet, but just in case...? */
+
+#define MALI_TILE_NO_FLAG (0)
+
+/* Helpers to generate tile coordinates based on the boundary coordinates in
+ * screen space. So, with the bounds (0, 0) to (128, 128) for the screen, these
+ * functions would convert it to the bounding tiles (0, 0) to (7, 7).
+ * Intentional "off-by-one"; finding the tile number is a form of fencepost
+ * problem. */
+
+#define MALI_MAKE_TILE_COORDS(X, Y) ((X) | ((Y) << 16))
+#define MALI_BOUND_TO_TILE(B, bias) ((B - bias) >> MALI_TILE_SHIFT)
+#define MALI_COORDINATE_TO_TILE(W, H, bias) MALI_MAKE_TILE_COORDS(MALI_BOUND_TO_TILE(W, bias), MALI_BOUND_TO_TILE(H, bias))
+#define MALI_COORDINATE_TO_TILE_MIN(W, H) MALI_COORDINATE_TO_TILE(W, H, 0)
+#define MALI_COORDINATE_TO_TILE_MAX(W, H) MALI_COORDINATE_TO_TILE(W, H, 1)
+
+struct mali_payload_fragment {
+ u32 min_tile_coord;
+ u32 max_tile_coord;
+ mali_ptr framebuffer;
+} __attribute__((packed));
+
+/* (Single?) Framebuffer Descriptor */
+
+/* Flags apply to format. With just MSAA_A and MSAA_B, the framebuffer is
+ * configured for 4x. With MSAA_8, it is configured for 8x. */
+
+#define MALI_FRAMEBUFFER_MSAA_8 (1 << 3)
+#define MALI_FRAMEBUFFER_MSAA_A (1 << 4)
+#define MALI_FRAMEBUFFER_MSAA_B (1 << 23)
+
+/* Fast/slow based on whether all three buffers are cleared at once */
+
+#define MALI_CLEAR_FAST (1 << 18)
+#define MALI_CLEAR_SLOW (1 << 28)
+#define MALI_CLEAR_SLOW_STENCIL (1 << 31)
+
+struct mali_single_framebuffer {
+ u32 unknown1;
+ u32 unknown2;
+ u64 unknown_address_0;
+ u64 zero1;
+ u64 zero0;
+
+ /* Exact format is ironically not known, since EGL is finnicky with the
+ * blob. MSAA, colourspace, etc are configured here. */
+
+ u32 format;
+
+ u32 clear_flags;
+ u32 zero2;
+
+ /* Purposeful off-by-one in these fields should be accounted for by the
+ * MALI_DIMENSION macro */
+
+ u16 width;
+ u16 height;
+
+ u32 zero3[8];
+
+ /* By default, the framebuffer is upside down from OpenGL's
+ * perspective. Set framebuffer to the end and negate the stride to
+ * flip in the Y direction */
+
+ mali_ptr framebuffer;
+ int32_t stride;
+
+ u32 zero4;
+
+ /* Depth and stencil buffers are interleaved, it appears, as they are
+ * set to the same address in captures. Both fields set to zero if the
+ * buffer is not being cleared. Depending on GL_ENABLE magic, you might
+ * get a zero enable despite the buffer being present; that still is
+ * disabled. */
+
+ mali_ptr depth_buffer; // not SAME_VA
+ u64 depth_buffer_enable;
+
+ mali_ptr stencil_buffer; // not SAME_VA
+ u64 stencil_buffer_enable;
+
+ u32 clear_color_1; // RGBA8888 from glClear, actually used by hardware
+ u32 clear_color_2; // always equal, but unclear function?
+ u32 clear_color_3; // always equal, but unclear function?
+ u32 clear_color_4; // always equal, but unclear function?
+
+ /* Set to zero if not cleared */
+
+ float clear_depth_1; // float32, ditto
+ float clear_depth_2; // float32, ditto
+ float clear_depth_3; // float32, ditto
+ float clear_depth_4; // float32, ditto
+
+ u32 clear_stencil; // Exactly as it appears in OpenGL
+
+ u32 zero6[7];
+
+ /* Very weird format, see generation code in trans_builder.c */
+ u32 resolution_check;
+
+ u32 tiler_flags;
+
+ u64 unknown_address_1; /* Pointing towards... a zero buffer? */
+ u64 unknown_address_2;
+
+ /* See mali_kbase_replay.c */
+ u64 tiler_heap_free;
+ u64 tiler_heap_end;
+
+ /* More below this, maybe */
+} __attribute__((packed));
+
+/* Format bits for the render target flags */
+
+#define MALI_MFBD_FORMAT_AFBC (1 << 5)
+#define MALI_MFBD_FORMAT_MSAA (1 << 7)
+
+struct mali_rt_format {
+ unsigned unk1 : 32;
+ unsigned unk2 : 3;
+
+ unsigned nr_channels : 2; /* MALI_POSITIVE */
+
+ unsigned flags : 11;
+
+ unsigned swizzle : 12;
+
+ unsigned unk4 : 4;
+} __attribute__((packed));
+
+struct bifrost_render_target {
+ struct mali_rt_format format;
+
+ u64 zero1;
+
+ union {
+ struct {
+ /* Stuff related to ARM Framebuffer Compression. When AFBC is enabled,
+ * there is an extra metadata buffer that contains 16 bytes per tile.
+ * The framebuffer needs to be the same size as before, since we don't
+ * know ahead of time how much space it will take up. The
+ * framebuffer_stride is set to 0, since the data isn't stored linearly
+ * anymore.
+ */
+
+ mali_ptr metadata;
+ u32 stride; // stride in units of tiles
+ u32 unk; // = 0x20000
+ } afbc;
+
+ struct {
+ /* Heck if I know */
+ u64 unk;
+ mali_ptr pointer;
+ } chunknown;
+ };
+
+ mali_ptr framebuffer;
+
+ u32 zero2 : 4;
+ u32 framebuffer_stride : 28; // in units of bytes
+ u32 zero3;
+
+ u32 clear_color_1; // RGBA8888 from glClear, actually used by hardware
+ u32 clear_color_2; // always equal, but unclear function?
+ u32 clear_color_3; // always equal, but unclear function?
+ u32 clear_color_4; // always equal, but unclear function?
+} __attribute__((packed));
+
+/* An optional part of bifrost_framebuffer. It comes between the main structure
+ * and the array of render targets. It must be included if any of these are
+ * enabled:
+ *
+ * - Transaction Elimination
+ * - Depth/stencil
+ * - TODO: Anything else?
+ */
+
+/* Flags field: note, these are guesses */
+
+#define MALI_EXTRA_PRESENT (0x400)
+#define MALI_EXTRA_AFBC (0x20)
+#define MALI_EXTRA_AFBC_ZS (0x10)
+#define MALI_EXTRA_ZS (0x4)
+
+struct bifrost_fb_extra {
+ mali_ptr checksum;
+ /* Each tile has an 8 byte checksum, so the stride is "width in tiles * 8" */
+ u32 checksum_stride;
+
+ u32 flags;
+
+ union {
+ /* Note: AFBC is only allowed for 24/8 combined depth/stencil. */
+ struct {
+ mali_ptr depth_stencil_afbc_metadata;
+ u32 depth_stencil_afbc_stride; // in units of tiles
+ u32 zero1;
+
+ mali_ptr depth_stencil;
+
+ u64 padding;
+ } ds_afbc;
+
+ struct {
+ /* Depth becomes depth/stencil in case of combined D/S */
+ mali_ptr depth;
+ u32 depth_stride_zero : 4;
+ u32 depth_stride : 28;
+ u32 zero1;
+
+ mali_ptr stencil;
+ u32 stencil_stride_zero : 4;
+ u32 stencil_stride : 28;
+ u32 zero2;
+ } ds_linear;
+ };
+
+
+ u64 zero3, zero4;
+} __attribute__((packed));
+
+/* flags for unk3 */
+
+/* Enables writing depth results back to main memory (rather than keeping them
+ * on-chip in the tile buffer and then discarding) */
+
+#define MALI_MFBD_DEPTH_WRITE (1 << 10)
+
+/* The MFBD contains the extra bifrost_fb_extra section */
+
+#define MALI_MFBD_EXTRA (1 << 13)
+
+struct bifrost_framebuffer {
+ u32 unk0; // = 0x10
+
+ u32 unknown2; // = 0x1f, same as SFBD
+ mali_ptr scratchpad;
+
+ /* 0x10 */
+ mali_ptr sample_locations;
+ mali_ptr unknown1;
+ /* 0x20 */
+ u16 width1, height1;
+ u32 zero3;
+ u16 width2, height2;
+ u32 unk1 : 19; // = 0x01000
+ u32 rt_count_1 : 2; // off-by-one (use MALI_POSITIVE)
+ u32 unk2 : 3; // = 0
+ u32 rt_count_2 : 3; // no off-by-one
+ u32 zero4 : 5;
+ /* 0x30 */
+ u32 clear_stencil : 8;
+ u32 unk3 : 24; // = 0x100
+ float clear_depth;
+ mali_ptr tiler_meta;
+ /* 0x40 */
+
+ /* Note: these are guesses! */
+ mali_ptr tiler_scratch_start;
+ mali_ptr tiler_scratch_middle;
+
+ /* These are not, since we see symmetry with replay jobs which name these explicitly */
+ mali_ptr tiler_heap_start;
+ mali_ptr tiler_heap_end;
+
+ u64 zero9, zero10, zero11, zero12;
+
+ /* optional: struct bifrost_fb_extra extra */
+ /* struct bifrost_render_target rts[] */
+} __attribute__((packed));
+
+#endif /* __PANFROST_JOB_H__ */
diff --git a/lib/panfrost-misc.h b/lib/panfrost-misc.h
new file mode 100644
index 000000000000..82363d8730bf
--- /dev/null
+++ b/lib/panfrost-misc.h
@@ -0,0 +1,47 @@
+/*
+ * © Copyright 2017-2018 The Panfrost Community
+ *
+ * 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.
+ *
+ */
+
+#ifndef __PANFROST_MISC_H__
+#define __PANFROST_MISC_H__
+
+#include <inttypes.h>
+
+typedef uint8_t u8;
+typedef uint16_t u16;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int8_t s8;
+typedef int16_t s16;
+typedef int32_t s32;
+typedef int64_t s64;
+
+typedef uint64_t mali_ptr;
+
+#define MALI_PTR_FMT "0x%" PRIx64
+
+/* FIXME: put this somewhere more fitting */
+#define MALI_MEM_MAP_TRACKING_HANDLE (3ull << 12)
+
+#endif
--
2.20.1
_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev
^ permalink raw reply related [flat|nested] 8+ messages in thread* [igt-dev] [PATCH i-g-t 5/6] tests/panfrost: Add initial tests for panfrost
2019-05-10 14:44 [igt-dev] [PATCH i-g-t 1/6] lib: Check for -ETIME, not ETIME Tomeu Vizoso
` (2 preceding siblings ...)
2019-05-10 14:44 ` [igt-dev] [PATCH i-g-t 4/6] lib: Add support for opening panfrost devices Tomeu Vizoso
@ 2019-05-10 14:44 ` Tomeu Vizoso
2019-05-10 14:44 ` [igt-dev] [PATCH i-g-t 6/6] panfrost: Don't check for automake support Tomeu Vizoso
` (2 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Tomeu Vizoso @ 2019-05-10 14:44 UTC (permalink / raw)
To: Development mailing list for IGT GPU Tools; +Cc: Petri Latvala, Tomeu Vizoso
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Acked-by: Petri Latvala <petri.latvala@intel.com>
---
tests/meson.build | 4 +
tests/panfrost_gem_new.c | 90 ++++++++++++++++++++
tests/panfrost_get_param.c | 73 +++++++++++++++++
tests/panfrost_prime.c | 79 ++++++++++++++++++
tests/panfrost_submit.c | 163 +++++++++++++++++++++++++++++++++++++
5 files changed, 409 insertions(+)
create mode 100644 tests/panfrost_gem_new.c
create mode 100644 tests/panfrost_get_param.c
create mode 100644 tests/panfrost_prime.c
create mode 100644 tests/panfrost_submit.c
diff --git a/tests/meson.build b/tests/meson.build
index 711979b4a1c2..351594fa72f0 100644
--- a/tests/meson.build
+++ b/tests/meson.build
@@ -63,6 +63,10 @@ test_progs = [
'kms_vblank',
'kms_vrr',
'meta_test',
+ 'panfrost_get_param',
+ 'panfrost_gem_new',
+ 'panfrost_prime',
+ 'panfrost_submit',
'perf',
'prime_busy',
'prime_mmap',
diff --git a/tests/panfrost_gem_new.c b/tests/panfrost_gem_new.c
new file mode 100644
index 000000000000..940525ff1b34
--- /dev/null
+++ b/tests/panfrost_gem_new.c
@@ -0,0 +1,90 @@
+/*
+ * Copyright © 2016 Broadcom
+ * Copyright © 2019 Collabora, Ltd.
+ *
+ * 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.
+ */
+
+#include "igt.h"
+#include "igt_panfrost.h"
+#include <unistd.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <fcntl.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <sys/stat.h>
+#include <sys/ioctl.h>
+#include "panfrost_drm.h"
+
+igt_main
+{
+ int fd;
+
+ igt_fixture {
+ fd = drm_open_driver(DRIVER_PANFROST);
+ }
+
+ igt_subtest("gem-new-4096") {
+ struct panfrost_bo *bo = igt_panfrost_gem_new(fd, 4096);
+ igt_panfrost_free_bo(fd, bo);
+ }
+
+ igt_subtest("gem-new-0") {
+ struct drm_panfrost_create_bo arg = {
+ .size = 0,
+ };
+ do_ioctl_err(fd, DRM_IOCTL_PANFROST_CREATE_BO, &arg, EINVAL);
+ }
+
+ igt_subtest("gem-new-zeroed") {
+ int fd2 = drm_open_driver(DRIVER_PANFROST);
+ struct panfrost_bo *bo;
+ uint32_t *map;
+ /* A size different from any used in our other tests, to try
+ * to convince it to land as the only one of its size in the
+ * kernel BO cache
+ */
+ size_t size = 3 * 4096, i;
+
+ /* Make a BO and free it on our main fd. */
+ bo = igt_panfrost_gem_new(fd, size);
+ map = igt_panfrost_mmap_bo(fd, bo->handle, size, PROT_READ | PROT_WRITE);
+ memset(map, 0xd0, size);
+ munmap(map, size);
+ igt_panfrost_free_bo(fd, bo);
+
+ /* Now, allocate a BO on the other fd and make sure it doesn't
+ * have the old contents.
+ */
+ bo = igt_panfrost_gem_new(fd2, size);
+ map = igt_panfrost_mmap_bo(fd2, bo->handle, size, PROT_READ | PROT_WRITE);
+ for (i = 0; i < size / 4; i++)
+ igt_assert_eq_u32(map[i], 0x0);
+ munmap(map, size);
+ igt_panfrost_free_bo(fd2, bo);
+
+ close(fd2);
+ }
+
+ igt_fixture
+ close(fd);
+}
diff --git a/tests/panfrost_get_param.c b/tests/panfrost_get_param.c
new file mode 100644
index 000000000000..11c2632b8f3b
--- /dev/null
+++ b/tests/panfrost_get_param.c
@@ -0,0 +1,73 @@
+/*
+ * Copyright © 2017 Broadcom
+ * Copyright © 2019 Collabora, Ltd.
+ *
+ * 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.
+ */
+
+#include "igt.h"
+#include "igt_panfrost.h"
+#include <unistd.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <fcntl.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <sys/stat.h>
+#include <sys/ioctl.h>
+#include <poll.h>
+#include "panfrost_drm.h"
+
+igt_main
+{
+ int fd;
+
+ igt_fixture
+ fd = drm_open_driver(DRIVER_PANFROST);
+
+ igt_subtest("base-params") {
+ int last_base_param = DRM_PANFROST_PARAM_GPU_PROD_ID;
+ uint32_t results[last_base_param + 1];
+
+ for (int i = 0; i < ARRAY_SIZE(results); i++)
+ results[i] = igt_panfrost_get_param(fd, i);
+
+ igt_assert(results[DRM_PANFROST_PARAM_GPU_PROD_ID]);
+ }
+
+ igt_subtest("get-bad-param") {
+ struct drm_panfrost_get_param get = {
+ .param = 0xd0d0d0d0,
+ };
+ do_ioctl_err(fd, DRM_IOCTL_PANFROST_GET_PARAM, &get, EINVAL);
+ }
+
+ igt_subtest("get-bad-padding") {
+ struct drm_panfrost_get_param get = {
+ .param = DRM_PANFROST_PARAM_GPU_PROD_ID,
+ .pad = 1,
+ };
+ do_ioctl_err(fd, DRM_IOCTL_PANFROST_GET_PARAM, &get, EINVAL);
+ }
+
+ igt_fixture
+ close(fd);
+}
diff --git a/tests/panfrost_prime.c b/tests/panfrost_prime.c
new file mode 100644
index 000000000000..351d46f2f7e6
--- /dev/null
+++ b/tests/panfrost_prime.c
@@ -0,0 +1,79 @@
+/*
+ * Copyright © 2016 Broadcom
+ * Copyright © 2019 Collabora, Ltd.
+ *
+ * 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.
+ */
+
+#include "igt.h"
+#include "igt_panfrost.h"
+#include <unistd.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <fcntl.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <sys/stat.h>
+#include <sys/ioctl.h>
+#include "panfrost_drm.h"
+
+igt_main
+{
+ int fd, kms_fd;
+
+ igt_fixture {
+ kms_fd = drm_open_driver_master(DRIVER_ANY);
+ fd = drm_open_driver(DRIVER_PANFROST);
+ }
+
+ igt_subtest("gem-prime-import") {
+ struct panfrost_bo *bo;
+ uint32_t handle, dumb_handle;
+ struct drm_panfrost_get_bo_offset get_bo_offset = {0,};
+ int dmabuf_fd;
+
+ /* Just to be sure that when we import the dumb buffer it has
+ * a non-NULL address.
+ */
+ bo = igt_panfrost_gem_new(fd, 1024);
+
+ dumb_handle = kmstest_dumb_create(kms_fd, 1024, 1024, 32, NULL, NULL);
+
+ dmabuf_fd = prime_handle_to_fd(kms_fd, dumb_handle);
+
+ handle = prime_fd_to_handle(fd, dmabuf_fd);
+
+ get_bo_offset.handle = handle;
+ do_ioctl(fd, DRM_IOCTL_PANFROST_GET_BO_OFFSET, &get_bo_offset);
+ igt_assert(get_bo_offset.offset);
+
+ gem_close(fd, handle);
+
+ kmstest_dumb_destroy(kms_fd, dumb_handle);
+
+ igt_panfrost_free_bo(fd, bo);
+ }
+
+ igt_fixture {
+ close(fd);
+ close(kms_fd);
+ }
+}
diff --git a/tests/panfrost_submit.c b/tests/panfrost_submit.c
new file mode 100644
index 000000000000..5770dc24a42b
--- /dev/null
+++ b/tests/panfrost_submit.c
@@ -0,0 +1,163 @@
+/*
+ * Copyright © 2016 Broadcom
+ * Copyright © 2019 Collabora, Ltd.
+ *
+ * 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.
+ */
+
+#include "igt.h"
+#include "igt_panfrost.h"
+#include "igt_syncobj.h"
+#include <unistd.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <fcntl.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <sys/stat.h>
+#include <sys/ioctl.h>
+#include "panfrost_drm.h"
+
+#define WIDTH 1366
+#define HEIGHT 768
+#define CLEAR_COLOR 0xff7f7f7f
+
+/* One tenth of a second */
+#define SHORT_TIME_NSEC 100000000ull
+
+/* Add the time that the bad job takes to timeout (sched->timeout) and the time that a reset can take */
+#define BAD_JOB_TIME_NSEC (SHORT_TIME_NSEC + 500000000ull + 100000000ull)
+
+#define NSECS_PER_SEC 1000000000ull
+
+static uint64_t
+abs_timeout(uint64_t duration)
+{
+ struct timespec current;
+ clock_gettime(CLOCK_MONOTONIC, ¤t);
+ return (uint64_t)current.tv_sec * NSECS_PER_SEC + current.tv_nsec + duration;
+}
+
+static void check_fb(int fd, struct panfrost_bo *bo)
+{
+ __uint32_t *fbo;
+ int i;
+
+ fbo = bo->map;
+ for (i = 0; i < ALIGN(WIDTH, 16) * HEIGHT; i++)
+ igt_assert_eq_u32(fbo[i], CLEAR_COLOR);
+}
+
+igt_main
+{
+ int fd;
+
+ igt_fixture {
+ fd = drm_open_driver(DRIVER_PANFROST);
+ }
+
+ igt_subtest("pan-submit") {
+ struct panfrost_submit *submit;
+
+ submit = igt_panfrost_trivial_job(fd, false, WIDTH, HEIGHT,
+ CLEAR_COLOR);
+
+ igt_panfrost_bo_mmap(fd, submit->fbo);
+ do_ioctl(fd, DRM_IOCTL_PANFROST_SUBMIT, submit->args);
+ igt_assert(syncobj_wait(fd, &submit->args->out_sync, 1,
+ abs_timeout(SHORT_TIME_NSEC), 0, NULL));
+ check_fb(fd, submit->fbo);
+ igt_panfrost_free_job(fd, submit);
+ }
+
+ igt_subtest("pan-submit-error-no-jc") {
+ struct drm_panfrost_submit submit = {.jc = 0,};
+ do_ioctl_err(fd, DRM_IOCTL_PANFROST_SUBMIT, &submit, EINVAL);
+ }
+
+ igt_subtest("pan-submit-error-bad-in-syncs") {
+ struct panfrost_submit *submit;
+
+ submit = igt_panfrost_trivial_job(fd, false, WIDTH, HEIGHT,
+ CLEAR_COLOR);
+ submit->args->in_syncs = 0ULL;
+ submit->args->in_sync_count = 1;
+
+ do_ioctl_err(fd, DRM_IOCTL_PANFROST_SUBMIT, submit->args, EFAULT);
+ }
+
+ igt_subtest("pan-submit-error-bad-bo-handles") {
+ struct panfrost_submit *submit;
+
+ submit = igt_panfrost_trivial_job(fd, false, WIDTH, HEIGHT,
+ CLEAR_COLOR);
+ submit->args->bo_handles = 0ULL;
+ submit->args->bo_handle_count = 1;
+
+ do_ioctl_err(fd, DRM_IOCTL_PANFROST_SUBMIT, submit->args, EFAULT);
+ }
+
+ igt_subtest("pan-submit-error-bad-requirements") {
+ struct panfrost_submit *submit;
+
+ submit = igt_panfrost_trivial_job(fd, false, WIDTH, HEIGHT,
+ CLEAR_COLOR);
+ submit->args->requirements = 2;
+
+ do_ioctl_err(fd, DRM_IOCTL_PANFROST_SUBMIT, submit->args, EINVAL);
+ }
+
+ igt_subtest("pan-submit-error-bad-out-sync") {
+ struct panfrost_submit *submit;
+
+ submit = igt_panfrost_trivial_job(fd, false, WIDTH, HEIGHT,
+ CLEAR_COLOR);
+ submit->args->out_sync = -1;
+
+ do_ioctl_err(fd, DRM_IOCTL_PANFROST_SUBMIT, submit->args, ENODEV);
+ }
+
+ igt_subtest("pan-reset") {
+ struct panfrost_submit *submit;
+
+ submit = igt_panfrost_trivial_job(fd, true, WIDTH, HEIGHT,
+ CLEAR_COLOR);
+ do_ioctl(fd, DRM_IOCTL_PANFROST_SUBMIT, submit->args);
+ /* Expect for this job to timeout */
+ igt_assert(!syncobj_wait(fd, &submit->args->out_sync, 1,
+ abs_timeout(SHORT_TIME_NSEC), 0, NULL));
+ igt_panfrost_free_job(fd, submit);
+
+ submit = igt_panfrost_trivial_job(fd, false, WIDTH, HEIGHT,
+ CLEAR_COLOR);
+ igt_panfrost_bo_mmap(fd, submit->fbo);
+ do_ioctl(fd, DRM_IOCTL_PANFROST_SUBMIT, submit->args);
+ /* This one should work */
+ igt_assert(syncobj_wait(fd, &submit->args->out_sync, 1,
+ abs_timeout(BAD_JOB_TIME_NSEC), 0, NULL));
+ check_fb(fd, submit->fbo);
+ igt_panfrost_free_job(fd, submit);
+ }
+
+ igt_fixture {
+ close(fd);
+ }
+}
--
2.20.1
_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev
^ permalink raw reply related [flat|nested] 8+ messages in thread