[igt-dev] [PATCH i-g-t v3 4/8] tests/i915/gem_blits: Use new copy instruction
Zbigniew Kempczyński
zbigniew.kempczynski at intel.com
Wed Mar 15 10:42:54 UTC 2023
On Tue, Mar 14, 2023 at 03:44:36PM +0100, Karolina Stolarek wrote:
> From: Arjun Melkaveri <arjun.melkaveri at intel.com>
>
> The test uses legacy command which is not supported on
> newer GPU generations. Use XY_FAST_COPY_BLT on newer GPU generations.
>
> Signed-off-by: Arjun Melkaveri <arjun.melkaveri at intel.com>
> Co-developed-by: Nirmoy Das <nirmoy.das at intel.com>
> Signed-off-by: Nirmoy Das <nirmoy.das at intel.com>
> Signed-off-by: Fei Yang <fei.yang at intel.com>
> Signed-off-by: Karolina Stolarek <karolina.stolarek at intel.com>
> Cc: Andi Shyti <andi.shyti at linux.intel.com>
> ---
> lib/intel_batchbuffer.c | 10 ++---
> lib/intel_batchbuffer.h | 6 +++
> tests/i915/gem_blits.c | 87 +++++++++++++++++++++++++----------------
> 3 files changed, 64 insertions(+), 39 deletions(-)
>
> diff --git a/lib/intel_batchbuffer.c b/lib/intel_batchbuffer.c
> index 7bb24c8f..a2bf5d2e 100644
> --- a/lib/intel_batchbuffer.c
> +++ b/lib/intel_batchbuffer.c
> @@ -92,8 +92,8 @@ static uint32_t fast_copy_pitch(unsigned int stride, unsigned int tiling)
> return stride;
> }
>
> -static uint32_t fast_copy_dword0(unsigned int src_tiling,
> - unsigned int dst_tiling)
> +uint32_t fast_copy_dword0(unsigned int src_tiling,
> + unsigned int dst_tiling)
> {
> uint32_t dword0 = 0;
>
> @@ -136,9 +136,9 @@ static uint32_t fast_copy_dword0(unsigned int src_tiling,
> return dword0;
> }
>
> -static uint32_t fast_copy_dword1(unsigned int src_tiling,
> - unsigned int dst_tiling,
> - int bpp)
> +uint32_t fast_copy_dword1(unsigned int src_tiling,
> + unsigned int dst_tiling,
> + int bpp)
> {
> uint32_t dword1 = 0;
>
> diff --git a/lib/intel_batchbuffer.h b/lib/intel_batchbuffer.h
> index 37db0ffa..81830d77 100644
> --- a/lib/intel_batchbuffer.h
> +++ b/lib/intel_batchbuffer.h
> @@ -31,6 +31,12 @@ enum i915_compression {
> I915_COMPRESSION_MEDIA,
> };
>
> +uint32_t fast_copy_dword0(unsigned int src_tiling,
> + unsigned int dst_tiling);
> +uint32_t fast_copy_dword1(unsigned int src_tiling,
> + unsigned int dst_tiling,
> + int bpp);
> +
> void igt_blitter_src_copy(int fd,
> uint64_t ahnd,
> uint32_t ctx,
> diff --git a/tests/i915/gem_blits.c b/tests/i915/gem_blits.c
> index 9ea3925c..e5ef3662 100644
> --- a/tests/i915/gem_blits.c
> +++ b/tests/i915/gem_blits.c
> @@ -22,15 +22,19 @@
> *
> */
>
> +#include "intel_batchbuffer.h"
> #include "i915/gem.h"
> #include "i915/gem_create.h"
> #include "igt.h"
> #include "igt_x86.h"
> +#include "i915/i915_blt.h"
>
> #define BCS_SWCTRL 0x22200
> #define BCS_SRC_Y (1 << 0)
> #define BCS_DST_Y (1 << 1)
>
> +static uint32_t devid;
> +
> struct device {
> int fd;
> int gen;
> @@ -145,8 +149,7 @@ static void buffer_set_tiling(const struct device *device,
> struct drm_i915_gem_relocation_entry reloc[2];
> struct drm_i915_gem_execbuffer2 execbuf;
> const bool has_64b_reloc = device->gen >= 8;
> - uint32_t stride, size, pitch;
> - uint32_t *batch;
> + uint32_t stride, size, pitch, *batch, dword1;
> int i;
>
> if (buffer->tiling == tiling)
> @@ -207,19 +210,25 @@ static void buffer_set_tiling(const struct device *device,
> batch[i++] = mask;
> }
>
> - batch[i] = (XY_SRC_COPY_BLT_CMD |
> - XY_SRC_COPY_BLT_WRITE_ALPHA |
> - XY_SRC_COPY_BLT_WRITE_RGB);
> - if (device->gen >= 4 && buffer->tiling)
> - batch[i] |= XY_SRC_COPY_BLT_SRC_TILED;
> - if (device->gen >= 4 && tiling)
> - batch[i] |= XY_SRC_COPY_BLT_DST_TILED;
> - batch[i++] |= 6 + 2 * has_64b_reloc;
> -
> pitch = stride;
> if (device->gen >= 4 && tiling)
> pitch /= 4;
> - batch[i++] = 3 << 24 | 0xcc << 16 | pitch;
> +
> + if (!blt_has_xy_src_copy(device->fd)) {
> + batch[i++] = fast_copy_dword0(buffer->tiling, tiling);
> + dword1 = fast_copy_dword1(buffer->tiling, tiling, 32);
> + batch[i++] = dword1 | pitch;
> + } else {
> + batch[i] = (XY_SRC_COPY_BLT_CMD |
> + XY_SRC_COPY_BLT_WRITE_ALPHA |
> + XY_SRC_COPY_BLT_WRITE_RGB);
> + if (device->gen >= 4 && buffer->tiling)
> + batch[i] |= XY_SRC_COPY_BLT_SRC_TILED;
> + if (device->gen >= 4 && tiling)
> + batch[i] |= XY_SRC_COPY_BLT_DST_TILED;
> + batch[i++] |= 6 + 2 * has_64b_reloc;
> + batch[i++] = 3 << 24 | 0xcc << 16 | pitch;
> + }
> batch[i++] = 0;
> batch[i++] = buffer->height << 16 | buffer->width;
> reloc[0].target_handle = obj[0].handle;
> @@ -296,8 +305,7 @@ static bool blit_to_linear(const struct device *device,
> struct drm_i915_gem_relocation_entry reloc[2];
> struct drm_i915_gem_execbuffer2 execbuf;
> const bool has_64b_reloc = device->gen >= 8;
> - uint32_t *batch;
> - uint32_t pitch;
> + uint32_t *batch, pitch, dword1;
> int i = 0;
>
> igt_assert(buffer->tiling);
> @@ -352,14 +360,19 @@ static bool blit_to_linear(const struct device *device,
> batch[i++] = mask;
> }
>
> - batch[i] = (XY_SRC_COPY_BLT_CMD |
> - XY_SRC_COPY_BLT_WRITE_ALPHA |
> - XY_SRC_COPY_BLT_WRITE_RGB);
> - if (device->gen >= 4 && buffer->tiling)
> - batch[i] |= XY_SRC_COPY_BLT_SRC_TILED;
> - batch[i++] |= 6 + 2 * has_64b_reloc;
> -
> - batch[i++] = 3 << 24 | 0xcc << 16 | buffer->stride;
> + if (!blt_has_xy_src_copy(device->fd)) {
> + batch[i++] = fast_copy_dword0(buffer->tiling, I915_TILING_NONE);
> + dword1 = fast_copy_dword1(buffer->tiling, I915_TILING_NONE, 32);
> + batch[i++] = dword1 | buffer->stride;
> + } else {
> + batch[i] = (XY_SRC_COPY_BLT_CMD |
> + XY_SRC_COPY_BLT_WRITE_ALPHA |
> + XY_SRC_COPY_BLT_WRITE_RGB);
> + if (device->gen >= 4 && buffer->tiling)
> + batch[i] |= XY_SRC_COPY_BLT_SRC_TILED;
> + batch[i++] |= 6 + 2 * has_64b_reloc;
> + batch[i++] = 3 << 24 | 0xcc << 16 | buffer->stride;
> + }
> batch[i++] = 0;
> batch[i++] = buffer->height << 16 | buffer->width;
> reloc[0].target_handle = obj[0].handle;
> @@ -598,8 +611,7 @@ blit(const struct device *device,
> struct drm_i915_gem_relocation_entry reloc[2];
> struct drm_i915_gem_execbuffer2 execbuf;
> const bool has_64b_reloc = device->gen >= 8;
> - uint32_t *batch;
> - uint32_t pitch;
> + uint32_t *batch, dword1, pitch;
> int i = 0;
>
> if (src_x < 0) {
> @@ -687,20 +699,25 @@ blit(const struct device *device,
> batch[i++] = mask;
> }
>
> - batch[i] = (XY_SRC_COPY_BLT_CMD |
> - XY_SRC_COPY_BLT_WRITE_ALPHA |
> - XY_SRC_COPY_BLT_WRITE_RGB);
> - if (device->gen >= 4 && src->tiling)
> - batch[i] |= XY_SRC_COPY_BLT_SRC_TILED;
> - if (device->gen >= 4 && dst->tiling)
> - batch[i] |= XY_SRC_COPY_BLT_DST_TILED;
> - batch[i++] |= 6 + 2 * has_64b_reloc;
> -
> pitch = dst->stride;
> if (device->gen >= 4 && dst->tiling)
> pitch /= 4;
> - batch[i++] = 3 << 24 | 0xcc << 16 | pitch;
>
> + if (!blt_has_xy_src_copy(device->fd)) {
> + batch[i++] = fast_copy_dword0(src->tiling, dst->tiling);
> + dword1 = fast_copy_dword1(src->tiling, dst->tiling, 32);
> + batch[i++] = dword1 | pitch;
> + } else {
> + batch[i] = (XY_SRC_COPY_BLT_CMD |
> + XY_SRC_COPY_BLT_WRITE_ALPHA |
> + XY_SRC_COPY_BLT_WRITE_RGB);
> + if (device->gen >= 4 && src->tiling)
> + batch[i] |= XY_SRC_COPY_BLT_SRC_TILED;
> + if (device->gen >= 4 && dst->tiling)
> + batch[i] |= XY_SRC_COPY_BLT_DST_TILED;
> + batch[i++] |= 6 + 2 * has_64b_reloc;
> + batch[i++] = 3 << 24 | 0xcc << 16 | pitch;
> + }
> batch[i++] = dst_y << 16 | dst_x;
> batch[i++] = (height + dst_y) << 16 | (width + dst_x);
> reloc[0].target_handle = obj[0].handle;
> @@ -792,6 +809,8 @@ igt_main
> struct buffer *src, *dst;
> unsigned int x, y;
>
> + devid = intel_get_drm_devid(device.fd);
Seems unused (+static on top).
With this fixed:
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski at intel.com>
--
Zbigniew
> +
> for (unsigned int height = 1; height <= 16; height <<= 2) {
> for (unsigned int y0 = ZERO; y0 <= (height > 2 ? BELOW : ZERO); y0++) {
> for (unsigned int width = 1; width <= 64; width <<= 2) {
> --
> 2.25.1
>
More information about the igt-dev
mailing list