[igt-dev] [PATCH i-g-t v2 1/2] lib/intel_bufops: Introduce buffer operations
Katarzyna Dec
katarzyna.dec at intel.com
Tue Dec 3 09:41:56 UTC 2019
On Tue, Dec 03, 2019 at 10:25:29AM +0100, Katarzyna Dec wrote:
> On Sun, Dec 01, 2019 at 08:09:41PM +0100, Zbigniew Kempczyński wrote:
> > Different GENs supports different tile surfaces. Older GENs
> > have HW fences to allow X / Y surface tiling / detiling.
> > Newer GENs have to tile / detile such surface in software.
> >
> > To make test developer life easier this code adds buffer
> > operations (short bufops) to use appropriate functions allowing
> > copying linear buffer to BO and from BO to linear buffer
> > regardless GPU generation and tiling within BO. For GENs having
> > fences support preference is to use them if they are available
> > (X / Y tiling is probed on bufops initalization).
> >
> > Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski at intel.com>
> > Cc: Chris Wilson <chris at chris-wilson.co.uk>
> > Cc: Imre Deak <imre.deak at intel.com>
> > Cc: Katarzyna Dec <katarzyna.dec at intel.com>
> > ---
> > lib/Makefile.sources | 2 +
> > lib/intel_bufops.c | 689 +++++++++++++++++++++++++++++++++++++++++++
> > lib/intel_bufops.h | 24 ++
> > lib/meson.build | 1 +
> > 4 files changed, 716 insertions(+)
> > create mode 100644 lib/intel_bufops.c
> > create mode 100644 lib/intel_bufops.h
> >
> > diff --git a/lib/Makefile.sources b/lib/Makefile.sources
> > index 9d1a4e06..b808f294 100644
> > --- a/lib/Makefile.sources
> > +++ b/lib/Makefile.sources
> > @@ -65,6 +65,8 @@ lib_source_list = \
> > intel_aub.h \
> > intel_batchbuffer.c \
> > intel_batchbuffer.h \
> > + intel_bufops.c \
> > + intel_bufops.h \
> > intel_chipset.c \
> > intel_chipset.h \
> > intel_device_info.c \
> > diff --git a/lib/intel_bufops.c b/lib/intel_bufops.c
> > new file mode 100644
> > index 00000000..fdb8bdc7
> > --- /dev/null
> > +++ b/lib/intel_bufops.c
> > @@ -0,0 +1,689 @@
> > +#include <sys/ioctl.h>
> > +#include "igt.h"
> > +#include "igt_x86.h"
> > +#include "intel_bufops.h"
> > +
> > +#define TILE_NONE (1 << I915_TILING_NONE)
> > +#define TILE_X (1 << I915_TILING_X)
> > +#define TILE_Y (1 << I915_TILING_Y)
> > +#define TILE_Yf (1 << I915_TILING_Yf)
> > +#define TILE_Ys (1 << I915_TILING_Ys)
> > +
> > +typedef void (*bo_copy)(int, struct igt_buf *, uint32_t *);
> > +
> > +struct buf_ops {
> > + int fd;
> > + drm_intel_bufmgr *bufmgr;
> > + int gen_start;
> > + int gen_end;
> > + int intel_gen;
> > + uint32_t supported_tiles;
> > + uint32_t hw_tiles;
> > + bo_copy linear_to;
> > + bo_copy linear_to_x;
> > + bo_copy linear_to_y;
> > + bo_copy linear_to_yf;
> > + bo_copy linear_to_ys;
> > + bo_copy to_linear;
> > + bo_copy x_to_linear;
> > + bo_copy y_to_linear;
> > + bo_copy yf_to_linear;
> > + bo_copy ys_to_linear;
> > +};
> > +
> > +static void *x_ptr(void *ptr,
> > + unsigned int x, unsigned int y,
> > + unsigned int stride, unsigned int cpp)
> > +{
> > + const int tile_width = 512;
> > + const int tile_height = 8;
> It can be really silly question - are these ^ values always the same?
> > + const int tile_size = tile_width * tile_height;
> > + int tile_x, tile_y;
> > + int offset_x, offset_y, pos;
> > +
> > + x *= cpp;
> > + tile_x = x / tile_width;
> > + tile_y = y / tile_height;
> > + offset_x = (tile_x * tile_size);
> > + offset_y = (tile_y * stride * tile_height);
> > +
> > + pos = offset_y + offset_x +
> > + (y % tile_height * tile_width) + (x % tile_width);
> > +
> > + return ptr + pos;
> > +}
> > +
> > +static void *y_ptr(void *ptr,
> > + unsigned int x, unsigned int y,
> > + unsigned int stride, unsigned int cpp)
> > +{
> > + const int tile_width = 128;
> > + const int tile_height = 32;
> > + const int owords = 16;
> Similar question. I guess this is something specific either to Intel Gen or to
> testcases.
> > + const int tile_size = tile_width * tile_height;
> > + int tile_x, tile_y;
> > + int offset_x, offset_y, pos;
> > + int shift_x, shift_y;
> > +
> > + x *= cpp;
> > + tile_x = x / tile_width;
> > + tile_y = y / tile_height;
> > + offset_x = tile_x * tile_size;
> > + offset_y = tile_y * stride * tile_height;
> > + shift_x = x % owords + (x % tile_width) / owords * tile_width * cpp;
> > + shift_y = y % tile_height * owords;
> > +
> > + pos = offset_y + offset_x + shift_x + shift_y;
> > +
> > + return ptr + pos;
> > +}
> > +
> > +static void *yf_ptr(void *ptr,
> > + unsigned int x, unsigned int y,
> > + unsigned int stride, unsigned int cpp)
> > +{
> > + const int tile_size = 4 * 1024;
> > + const int tile_width = 128;
> > + int row_size = (stride / tile_width) * tile_size;
> > +
> > + x *= cpp; /* convert to Byte offset */
> > +
> > +
> > + /*
> > + * Within a 4k Yf tile, the byte swizzling pattern is
> > + * msb......lsb
> > + * xyxyxyyyxxxx
> > + * The tiles themselves are laid out in row major order.
> > + */
> > + return ptr +
> > + ((x & 0xf) * 1) + /* 4x1 pixels(32bpp) = 16B */
> > + ((y & 0x3) * 16) + /* 4x4 pixels = 64B */
> > + (((y & 0x4) >> 2) * 64) + /* 1x2 64B blocks */
> > + (((x & 0x10) >> 4) * 128) + /* 2x2 64B blocks = 256B block */
> > + (((y & 0x8) >> 3) * 256) + /* 2x1 256B blocks */
> > + (((x & 0x20) >> 5) * 512) + /* 2x2 256B blocks */
> > + (((y & 0x10) >> 4) * 1024) + /* 4x2 256 blocks */
> > + (((x & 0x40) >> 6) * 2048) + /* 4x4 256B blocks = 4k tile */
> > + (((x & ~0x7f) >> 7) * tile_size) + /* row of tiles */
> > + (((y & ~0x1f) >> 5) * row_size);
> > +}
> > +
> > +typedef void *(*fn_ptr)(void *, unsigned int, unsigned int,
> > + unsigned int, unsigned int);
> > +static fn_ptr __get_tile_fn_ptr(int tiling)
> > +{
> > + fn_ptr fn = NULL;
> > +
> > + switch (tiling) {
> > + case I915_TILING_X:
> > + fn = x_ptr;
> > + break;
> > + case I915_TILING_Y:
> > + fn = y_ptr;
> > + break;
> > + case I915_TILING_Yf:
> > + fn = yf_ptr;
> > + break;
> > + case I915_TILING_Ys:
> > + /* To be implemented */
> > + break;
> > + }
> > +
> > + igt_assert_f(fn, "Can't find tile function for tiling: %d\n", tiling);
> > +
> > + return fn;
> > +}
> > +
> > +static void __copy_linear_to(int fd, struct igt_buf *buf,
> > + const uint32_t *linear, int tiling)
> > +{
> > + int height = igt_buf_height(buf);
> > + int width = igt_buf_width(buf);
> > + fn_ptr fn = __get_tile_fn_ptr(tiling);
> > + void *map;
> > +
> > + gem_set_domain(fd, buf->bo->handle,
> > + I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
> > + map = gem_mmap__cpu(fd, buf->bo->handle, 0,
> > + buf->bo->size, PROT_READ | PROT_WRITE);
> > +
> > + for (int y = 0; y < height; y++) {
> > + for (int x = 0; x < width; x++) {
> > + uint32_t *ptr = fn(map, x, y,
> > + buf->stride, buf->bpp / 8);
> > +
> > + *ptr = linear[y * width + x];
> > + }
> > + }
> > +
> > + munmap(map, buf->bo->size);
> > +}
> > +
> > +static void copy_linear_to_x(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + __copy_linear_to(fd, buf, linear, I915_TILING_X);
> > +}
> > +
> > +static void copy_linear_to_y(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + __copy_linear_to(fd, buf, linear, I915_TILING_Y);
> > +}
> > +
> > +static void copy_linear_to_yf(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + __copy_linear_to(fd, buf, linear, I915_TILING_Yf);
> > +}
> > +
> > +static void __copy_to_linear(int fd, struct igt_buf *buf,
> > + uint32_t *linear, int tiling)
> > +{
> > + int height = igt_buf_height(buf);
> > + int width = igt_buf_width(buf);
> > + fn_ptr fn = __get_tile_fn_ptr(tiling);
> > + void *map;
> > +
> > + gem_set_domain(fd, buf->bo->handle,
> > + I915_GEM_DOMAIN_CPU, 0);
> > + map = gem_mmap__cpu(fd, buf->bo->handle, 0,
> > + buf->bo->size, PROT_READ);
> > +
> > + for (int y = 0; y < height; y++) {
> > + for (int x = 0; x < width; x++) {
> > + uint32_t *ptr = fn(map, x, y,
> > + buf->stride, buf->bpp / 8);
> > +
> > + linear[y * width + x] = *ptr;
> > + }
> > + }
> > +
> > + munmap(map, buf->bo->size);
> > +}
> > +
> > +static void copy_yf_to_linear(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + __copy_to_linear(fd, buf, linear, I915_TILING_Yf);
> > +}
> > +
> > +static void copy_x_to_linear(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + __copy_to_linear(fd, buf, linear, I915_TILING_X);
> > +}
> > +
> > +static void copy_y_to_linear(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + __copy_to_linear(fd, buf, linear, I915_TILING_Y);
> > +}
> > +
> > +static void copy_linear_to_gtt(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + void *map;
> > +
> > + gem_set_domain(fd, buf->bo->handle,
> > + I915_GEM_DOMAIN_GTT, I915_GEM_DOMAIN_GTT);
> > +
> > + map = gem_mmap__gtt(fd, buf->bo->handle,
> > + buf->bo->size, PROT_READ | PROT_WRITE);
> > +
> > + memcpy(map, linear, buf->bo->size);
> > +
> > + munmap(map, buf->bo->size);
> > +}
> > +
> > +static void copy_gtt_to_linear(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + void *map;
> > +
> > + gem_set_domain(fd, buf->bo->handle,
> > + I915_GEM_DOMAIN_GTT, 0);
> > +
> > + map = gem_mmap__gtt(fd, buf->bo->handle,
> > + buf->bo->size, PROT_READ);
> > +
> > + igt_memcpy_from_wc(linear, map, buf->bo->size);
> > +
> > + munmap(map, buf->bo->size);
> > +}
> > +
> > +static void copy_linear_to_wc(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + void *map;
> > +
> > + gem_set_domain(fd, buf->bo->handle,
> > + I915_GEM_DOMAIN_GTT, I915_GEM_DOMAIN_GTT);
> > +
> > + map = gem_mmap__wc(fd, buf->bo->handle, 0,
> > + buf->bo->size, PROT_READ | PROT_WRITE);
> > +
> > + memcpy(map, linear, buf->bo->size);
> > +
> > + munmap(map, buf->bo->size);
> > +}
> > +
> > +static void copy_wc_to_linear(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + void *map;
> > +
> > + gem_set_domain(fd, buf->bo->handle,
> > + I915_GEM_DOMAIN_GTT, 0);
> > +
> > + map = gem_mmap__wc(fd, buf->bo->handle, 0,
> > + buf->bo->size, PROT_READ);
> > +
> > + igt_memcpy_from_wc(linear, map, buf->bo->size);
> > +
> > + munmap(map, buf->bo->size);
> > +}
> > +
> > +static void copy_linear_to_cpu(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + void *map;
> > +
> > + gem_set_domain(fd, buf->bo->handle,
> > + I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
> > +
> > + map = gem_mmap__cpu(fd, buf->bo->handle, 0,
> > + buf->bo->size, PROT_READ | PROT_WRITE);
> > +
> > + memcpy(map, linear, buf->bo->size);
> > +
> > + munmap(map, buf->bo->size);
> > +}
> > +
> > +static void copy_cpu_to_linear(int fd, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + void *map;
> > +
> > + gem_set_domain(fd, buf->bo->handle,
> > + I915_GEM_DOMAIN_CPU, 0);
> > +
> > + map = gem_mmap__cpu(fd, buf->bo->handle, 0,
> > + buf->bo->size, PROT_READ);
> > +
> > + memcpy(linear, map, buf->bo->size);
> > +
> > + munmap(map, buf->bo->size);
> > +}
> > +
> > +void copy_buf_to_linear(struct buf_ops *bops, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + igt_assert(bops);
> > +
> > + switch (buf->tiling) {
> > + case I915_TILING_NONE:
> > + igt_assert(bops->to_linear);
> > + bops->to_linear(bops->fd, buf, linear);
> > + break;
> > + case I915_TILING_X:
> > + igt_assert(bops->x_to_linear);
> > + bops->x_to_linear(bops->fd, buf, linear);
> > + break;
> > + case I915_TILING_Y:
> > + igt_assert(bops->y_to_linear);
> > + bops->y_to_linear(bops->fd, buf, linear);
> > + break;
> > + case I915_TILING_Yf:
> > + igt_assert(bops->yf_to_linear);
> > + bops->yf_to_linear(bops->fd, buf, linear);
> > + break;
> > + case I915_TILING_Ys:
> > + igt_assert(bops->ys_to_linear);
> > + bops->ys_to_linear(bops->fd, buf, linear);
> > + break;
> > + }
> > +}
> > +
> > +void copy_linear_to_buf(struct buf_ops *bops, struct igt_buf *buf,
> > + uint32_t *linear)
> > +{
> > + igt_assert(bops);
> > +
> > + switch (buf->tiling) {
> > + case I915_TILING_NONE:
> > + igt_assert(bops->linear_to);
> > + bops->linear_to(bops->fd, buf, linear);
> > + break;
> > + case I915_TILING_X:
> > + igt_assert(bops->linear_to_x);
> > + bops->linear_to_x(bops->fd, buf, linear);
> > + break;
> > + case I915_TILING_Y:
> > + igt_assert(bops->linear_to_y);
> > + bops->linear_to_y(bops->fd, buf, linear);
> > + break;
> > + case I915_TILING_Yf:
> > + igt_assert(bops->linear_to_yf);
> > + bops->linear_to_yf(bops->fd, buf, linear);
> > + break;
> > + case I915_TILING_Ys:
> > + igt_assert(bops->linear_to_ys);
> > + bops->linear_to_ys(bops->fd, buf, linear);
> > + break;
> > + }
> > +}
> > +
> > +struct buf_ops buf_ops_arr[] = {
> > + /* Generations 0 - 8 */
> > +{
> > + .gen_start = 0,
> > + .gen_end = 8,
> > + .supported_tiles = TILE_NONE | TILE_X | TILE_Y,
> > + .hw_tiles = TILE_X | TILE_Y,
> > + .linear_to = copy_linear_to_cpu,
> > + .linear_to_x = copy_linear_to_gtt,
> > + .linear_to_y = copy_linear_to_gtt,
> > + .to_linear = copy_cpu_to_linear,
> > + .x_to_linear = copy_gtt_to_linear,
> > + .y_to_linear = copy_gtt_to_linear,
> > +},
> > +/* Generations 9 - 11 */
> > +{
> > + .gen_start = 9,
> > + .gen_end = 11,
> > + .supported_tiles = TILE_NONE | TILE_X | TILE_Y | TILE_Yf,
> > + .hw_tiles = TILE_X | TILE_Y,
> > + .linear_to = copy_linear_to_cpu,
> > + .linear_to_x = copy_linear_to_gtt,
> > + .linear_to_y = copy_linear_to_gtt,
> > + .linear_to_yf = copy_linear_to_yf,
> > + .to_linear = copy_cpu_to_linear,
> > + .x_to_linear = copy_gtt_to_linear,
> > + .y_to_linear = copy_gtt_to_linear,
> > + .yf_to_linear = copy_yf_to_linear,
> > +},
> > +/* Generation 12 */
> > +{
> > + .gen_start = 12,
> > + .gen_end = 12,
> > + .supported_tiles = TILE_NONE | TILE_X | TILE_Y | TILE_Yf | TILE_Ys,
> > + .linear_to = copy_linear_to_wc,
> > + .linear_to_x = copy_linear_to_x,
> > + .linear_to_y = copy_linear_to_y,
> > + .linear_to_yf = copy_linear_to_yf,
> > + .linear_to_ys = NULL, /* to be implemented */
> > + .to_linear = copy_wc_to_linear,
> > + .x_to_linear = copy_x_to_linear,
> > + .y_to_linear = copy_y_to_linear,
> > + .yf_to_linear = copy_yf_to_linear,
> > + .ys_to_linear = NULL, /* to be implemented */
> > +},
> > +};
> > +
> > +static int __gem_get_tiling(int fd, struct drm_i915_gem_get_tiling *arg)
> > +{
> > + int err;
> > +
> > + err = 0;
> > + if (igt_ioctl(fd, DRM_IOCTL_I915_GEM_GET_TILING, arg))
> > + err = -errno;
> > + errno = 0;
> > +
> > + return err;
> > +}
> > +
> > +static bool __get_tiling(int fd, uint32_t handle, uint32_t *tiling,
> > + uint32_t *swizzle)
> > +{
> > + struct drm_i915_gem_get_tiling get_tiling;
> > +
> > + memset(&get_tiling, 0, sizeof(get_tiling));
> > + get_tiling.handle = handle;
> > +
> > + if (__gem_get_tiling(fd, &get_tiling) != 0)
> > + return false;
> > +
> > + *tiling = get_tiling.tiling_mode;
> > + *swizzle = get_tiling.swizzle_mode;
> > +
> > + return get_tiling.phys_swizzle_mode == get_tiling.swizzle_mode;
> > +}
> > +
> > +static bool probe_hw_tiling(int fd, uint32_t tiling)
> > +{
> > + struct drm_i915_gem_set_tiling st;
> > + uint64_t size = 1024 * 1024;
> > + uint32_t handle, buf_tiling, buf_swizzle;
> > + uint32_t stride;
> > + int ret;
> > + bool is_set;
> > +
> > + if (tiling == I915_TILING_X)
> > + stride = 512;
> > + else if (tiling == I915_TILING_Y)
> > + stride = 128;
> > + else
> > + return false;
> > +
> > + handle = gem_create(fd, size);
> > +
> > + /* Single shot, if no fences are available we fail immediately */
> > + st.handle = handle;
> > + st.tiling_mode = tiling;
> > + st.stride = tiling ? stride : 0;
> > + ret = ioctl(fd, DRM_IOCTL_I915_GEM_SET_TILING, &st);
> > + if (ret)
> > + goto end;
> > +
> > + is_set = __get_tiling(fd, handle, &buf_tiling, &buf_swizzle);
> > +
> > +end:
> > + gem_close(fd, handle);
> > +
> > + return is_set;
> > +}
> > +
> > +struct buf_ops *buf_ops_get(int fd, drm_intel_bufmgr *bufmgr)
> > +{
> > + struct buf_ops *bops = calloc(1, sizeof(*bops));
> > + uint32_t devid;
> > + int generation;
> > +
> > + igt_assert(bufmgr);
> > + igt_assert(bops);
> > +
> > + devid = intel_get_drm_devid(fd);
> > + generation = intel_gen(devid);
> > +
> > + /* Predefined settings */
> > + for (int i = 0; i < ARRAY_SIZE(buf_ops_arr); i++) {
> > + if (generation >= buf_ops_arr[i].gen_start &&
> > + generation <= buf_ops_arr[i].gen_end) {
> > + memcpy(bops, &buf_ops_arr[i], sizeof(*bops));
> > + bops->fd = fd;
> > + bops->bufmgr = bufmgr;
> > + bops->intel_gen = generation;
> > + igt_debug("generation: %d, supported tiles: 0x%02x\n",
> > + generation, bops->supported_tiles);
> > + break;
> > + }
> > + }
> > +
> > + /* Let's probe X and Y hw tiling support */
> > + if (bops->hw_tiles & TILE_X) {
> > + bool supported = probe_hw_tiling(fd, I915_TILING_X);
> > +
> > + igt_debug("X fence support: %s\n", supported ? "yes" : "no");
> > + if (!supported) {
> > + bops->hw_tiles &= ~TILE_X;
> > + bops->linear_to_x = copy_linear_to_x;
> > + bops->x_to_linear = copy_x_to_linear;
> > + }
> > + }
> > +
> > + if (bops->hw_tiles & TILE_Y) {
> > + bool supported = probe_hw_tiling(fd, I915_TILING_Y);
> > +
> > + igt_debug("Y fence support: %s\n", supported ? "yes" : "no");
> > + if (!supported) {
> > + bops->hw_tiles &= ~TILE_Y;
> > + bops->linear_to_y = copy_linear_to_y;
> > + bops->y_to_linear = copy_y_to_linear;
> > + }
> > + }
> > +
> > + igt_assert(bops->intel_gen);
> > +
> > + return bops;
> > +}
> > +
> > +void buf_ops_switch_to_sofware_tiling(struct buf_ops *bops)
> so far so good :) first typo ^ s/sofware/software/
> Check declaration too :)
> > +{
> > + igt_debug("Switch to software tiling/detiling of X/Y surfaces\n");
> > +
> > + igt_assert(bops);
> > +
> > + bops->hw_tiles = 0;
> > + bops->linear_to_x = copy_linear_to_x;
> > + bops->linear_to_y = copy_linear_to_y;
> > + bops->x_to_linear = copy_x_to_linear;
> > + bops->y_to_linear = copy_y_to_linear;
> > +}
> > +
> > +bool buf_ops_has_hw_fence(struct buf_ops *bops, uint32_t tiling)
> > +{
> > + uint32_t tile_mask = (1 << tiling);
> > +
> > + igt_assert(bops);
> > +
> > + if (tile_mask & bops->hw_tiles)
> > + return true;
> > +
> > + return false;
> > +}
> > +
> > +bool buf_ops_has_tiling_support(struct buf_ops *bops, uint32_t tiling)
> > +{
> > + uint32_t tile_mask = (1 << tiling);
> > +
> > + igt_assert(bops);
> > +
> > + if (tile_mask & bops->supported_tiles)
> > + return true;
> > +
> > + return false;
> > +}
> > +
> > +static int buf_aux_width(int gen, const struct igt_buf *buf)
Similar fuctions (buf_aux_height) as well are still in
tests/i915/gem_render_copy . Is it by purpose?
And it looks like these 2 ^ v are only used in lib while could be used in test
functions.
BTW - finally we have someone who wants to make life easier with rendercopy.
Kasia :)
> > +{
> > + /*
> > + * GEN12+: The AUX CCS unit size is 64 bytes mapping 4 main surface
> > + * tiles. Thus the width of the CCS unit is 4*32=128 pixels on the
> > + * main surface.
> > + */
> > + if (gen >= 12)
> > + return DIV_ROUND_UP(igt_buf_width(buf), 128) * 64;
> > +
> > + return DIV_ROUND_UP(igt_buf_width(buf), 1024) * 128;
> > +}
> > +
> > +static int buf_aux_height(int gen, const struct igt_buf *buf)
> > +{
> > + /*
> > + * GEN12+: The AUX CCS unit size is 64 bytes mapping 4 main surface
> > + * tiles. Thus the height of the CCS unit is 32 pixel rows on the main
> > + * surface.
> > + */
> > + if (gen >= 12)
> > + return DIV_ROUND_UP(igt_buf_height(buf), 32);
> > +
> > + return DIV_ROUND_UP(igt_buf_height(buf), 512) * 32;
> > +}
> > +
> > +void igt_buf_init(struct buf_ops *bops, struct igt_buf *buf,
> > + int width, int height, int bpp,
> > + uint32_t req_tiling, bool ccs)
> > +{
> > + uint32_t tiling = req_tiling;
> > + bool supports_hw_fences;
> > +
> > + igt_assert(bops);
> > +
> > + supports_hw_fences = buf_ops_has_hw_fence(bops, req_tiling);
> > +
> > + memset(buf, 0, sizeof(*buf));
> > +
> > + if (ccs) {
> > + int aux_width, aux_height;
> > + int size;
> > +
> > + igt_require(bops->intel_gen >= 9);
> > + igt_assert(req_tiling == I915_TILING_Y ||
> > + req_tiling == I915_TILING_Yf);
> > +
> > + /*
> > + * On GEN12+ we align the main surface to 4 * 4 main surface
> > + * tiles, which is 64kB. These 16 tiles are mapped by 4 AUX
> > + * CCS units, that is 4 * 64 bytes. These 4 CCS units are in
> > + * turn mapped by one L1 AUX page table entry.
> > + */
> > + if (bops->intel_gen >= 12)
> > + buf->stride = ALIGN(width * (bpp / 8), 128 * 4);
> > + else
> > + buf->stride = ALIGN(width * (bpp / 8), 128);
> > +
> > + if (bops->intel_gen >= 12)
> > + height = ALIGN(height, 4 * 32);
> > +
> > + buf->size = buf->stride * height;
> > + buf->tiling = tiling;
> > + buf->bpp = bpp;
> > +
> > + aux_width = buf_aux_width(bops->intel_gen, buf);
> > + aux_height = buf_aux_height(bops->intel_gen, buf);
> > +
> > + buf->aux.offset = buf->stride * ALIGN(height, 32);
> > + buf->aux.stride = aux_width;
> > +
> > + size = buf->aux.offset + aux_width * aux_height;
> > +
> > + buf->bo = drm_intel_bo_alloc(bops->bufmgr, "", size, 4096);
> > +
> > + if (tiling == I915_TILING_Y && supports_hw_fences) {
> > + drm_intel_bo_set_tiling(buf->bo, &tiling, buf->stride);
> > + igt_assert_eq(tiling, req_tiling);
> > + }
> > + } else {
> > + /* Requested bo must be hw tiled */
> > + if (supports_hw_fences) {
> > + unsigned long pitch;
> > +
> > + buf->bo = drm_intel_bo_alloc_tiled(bops->bufmgr, "",
> > + width, height,
> > + bpp / 8,
> > + &tiling, &pitch, 0);
> > + igt_assert_eq(tiling, req_tiling);
> > + igt_assert(buf->bo);
> > +
> > + buf->stride = pitch;
> > + buf->tiling = tiling;
> > + buf->size = pitch * height;
> > + buf->bpp = bpp;
> > + } else {
> > + /* Use bo alloc and software tiling/detiling */
> > + int size;
> > +
> > + buf->stride = ALIGN(width * (bpp / 8), 128);
> > + buf->size = buf->stride * height;
> > + buf->tiling = tiling;
> > + buf->bpp = bpp;
> > +
> > + size = buf->stride * ALIGN(height, 32);
> > +
> > + buf->bo = drm_intel_bo_alloc(bops->bufmgr, "", size,
> > + 4096);
> > + igt_assert(buf->bo);
> > + }
> > + }
> > +}
> > diff --git a/lib/intel_bufops.h b/lib/intel_bufops.h
> > new file mode 100644
> > index 00000000..e53c6ebc
> > --- /dev/null
> > +++ b/lib/intel_bufops.h
> > @@ -0,0 +1,24 @@
> > +#ifndef __INTEL_BUFOPS_H__
> > +#define __INTEL_BUFOPS_H__
> > +
> > +#include "intel_batchbuffer.h"
> > +
> > +struct buf_ops;
> > +
> > +struct buf_ops *buf_ops_get(int fd, drm_intel_bufmgr *bufmgr);
> > +void buf_ops_switch_to_sofware_tiling(struct buf_ops *bops);
> > +
> > +void copy_buf_to_linear(struct buf_ops *bops, struct igt_buf *buf,
> > + uint32_t *linear);
> > +
> > +void copy_linear_to_buf(struct buf_ops *bops, struct igt_buf *buf,
> > + uint32_t *linear);
> > +
> > +bool buf_ops_has_hw_fence(struct buf_ops *bops, uint32_t tiling);
> > +bool buf_ops_has_tiling_support(struct buf_ops *bops, uint32_t tiling);
> > +
> > +void igt_buf_init(struct buf_ops *bops, struct igt_buf *buf,
> > + int width, int height, int bpp,
> > + uint32_t req_tiling, bool ccs);
> > +
> > +#endif
> > diff --git a/lib/meson.build b/lib/meson.build
> > index 3f908912..a88c3e89 100644
> > --- a/lib/meson.build
> > +++ b/lib/meson.build
> > @@ -26,6 +26,7 @@ lib_sources = [
> > 'igt_x86.c',
> > 'instdone.c',
> > 'intel_batchbuffer.c',
> > + 'intel_bufops.c',
> > 'intel_chipset.c',
> > 'intel_device_info.c',
> > 'intel_os.c',
> > --
> > 2.23.0
> >
> Huge and quite important functions introduced!
> Kasia
> _______________________________________________
> igt-dev mailing list
> igt-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/igt-dev
More information about the igt-dev
mailing list