intel/blorp: Redescribe surfaces for copies

When copying data between two surfaces, independently increase the size
of each surface's format (bits-per-pixel) as alignment constraints
allow. Adjust the other surface parameters and blorp_copy() parameters
accordingly.

This fixes copies between the 16bpp YCRCB formats and 32bpp formats:
dEQP-VK.ycbcr.single_plane_copy.linear.linear.r8g8b8a8_to_g8b8g8r8_422
This new test failure was reported by Iván Briano.

More generally, this increases the efficiency of our copies. As shown in
the configuration pages of the PRMs, our sampler is able to fetch texels
at a fixed rate of texels / clock regardless of the texel size
(presumably our rendering hardware has similar behavior). By using the
largest texel size possible, we can transfer more bits / clock.

Improves the performance of a number of traces in the performance CI for
BMG:

* TotalWarWarhammer3 +2.24%
* Payday3 	     +1.87%
* BaldursGate3 	     +1.34%
* Control 	     +1.25%
* TotalWarPharaoh    +1.22%

Four additional traces are helped between +0.44% and +0.96%.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39974>

src/intel/blorp/blorp_blit.c

@@ -1811,6 +1811,12 @@ blorp_surf_convert_to_single_level_tile(const struct isl_device *isl_dev,
    if (info->aux_usage == ISL_AUX_USAGE_NONE)
       usage |= ISL_SURF_USAGE_DISABLE_AUX_BIT;
 
+   /* Aux-tt alignment only applies to the beginning of the resource. We
+    * might be pointing to some other subresource however.
+    */
+   if (offset_B > 0)
+      usage |= ISL_SURF_USAGE_NO_AUX_TT_ALIGNMENT_BIT;
+
    struct isl_surf *scaled_surf = &info->surf;
    struct isl_view *scaled_view = &info->view;
    bool ok UNUSED;
@@ -3070,6 +3076,151 @@ blorp_copy_get_formats(const struct isl_device *isl_dev,
    }
 }
 
+static int
+get_max_format_scale(const struct isl_device *isl_dev,
+                     const struct blorp_surface_info *info,
+                     uint32_t x, uint32_t width, uint32_t height)
+{
+   const bool full_width = u_minify(info->surf.logical_level0_px.width,
+                                    info->view.base_level) == width;
+   const bool full_height = u_minify(info->surf.logical_level0_px.height,
+                                     info->view.base_level) == height;
+
+   if (info->aux_usage != ISL_AUX_USAGE_NONE) {
+      /* CCS_D, MCS and HIZ don't support changing the format bpb. FCV_CCS_E
+       * could be supported, but it requires more collaboration between BLORP
+       * and drivers.
+       */
+      if (info->aux_usage != ISL_AUX_USAGE_CCS_E)
+         return 1;
+
+      /* CCS_E on gfx9-11 requires the surface's bpc not change. */
+      if (isl_dev->info->ver <= 11)
+         return 1;
+
+      /* On gfx12, CCS_E can survive a change in the format bpb. However, the
+       * RenderCompressionFormat must not change.
+       */
+      if (isl_dev->info->ver == 12) {
+         if (info->view.usage & (ISL_SURF_USAGE_RENDER_TARGET_BIT |
+                                 ISL_SURF_USAGE_STORAGE_BIT)) {
+            /* For some destinations, the clear color can be ignored only if
+             * the entire slice is covered.
+             */
+            if (!full_width || !full_height)
+               return 1;
+         } else if (info->view.usage & ISL_SURF_USAGE_TEXTURE_BIT) {
+            /* For textures, a replicated pixel must have been provided. */
+            if (!info->has_replicated_pixel)
+               return 1;
+         }
+      }
+   }
+
+   /* We don't support depth/stencil. */
+   if (isl_surf_usage_is_depth_or_stencil(info->surf.usage))
+      return 1;
+
+   /* We don't support NPOT formats */
+   const struct isl_format_layout *surf_fmtl =
+      isl_format_get_layout(info->surf.format);
+   if (surf_fmtl->bpb % 3 == 0)
+      return 1;
+
+   struct isl_tile_info surf_tile_info;
+   isl_surf_get_tile_info(&info->surf, &surf_tile_info);
+   uint32_t lod1_w = u_minify(info->surf.logical_level0_px.width, 1);
+   uint32_t phys_lod1_w = align(lod1_w, info->surf.image_alignment_el.w);
+
+   /* Find the format size which satisfies alignment requirements. */
+   for (int max_bpb = 128; max_bpb >= surf_fmtl->bpb; max_bpb /= 2) {
+      if (info->view.base_level >= 1 &&
+          phys_lod1_w * surf_fmtl->bpb % max_bpb)
+         continue;
+
+      if (x * surf_fmtl->bpb % max_bpb)
+         continue;
+
+      if (info->tile_x_sa * surf_fmtl->bpb % max_bpb)
+         continue;
+
+      if (width * surf_fmtl->bpb % max_bpb) {
+         /* For buffers/linear surfaces, don't ignore the width. Doing so may
+          * lead to accessing buffer memory out of bounds.
+          */
+         if (info->surf.tiling == ISL_TILING_LINEAR)
+            continue;
+
+         /* Partial width copies must be aligned to avoid stomping on
+          * neighboring pixels.
+          */
+         if (!full_width)
+            continue;
+
+         /* No need to scale the format if we'd only add more padding. */
+         if (width * surf_fmtl->bpb < max_bpb)
+            continue;
+      }
+
+      if (!(info->view.usage & ISL_SURF_USAGE_TEXTURE_BIT)) {
+         /* All surface types except for textures need their row pitch aligned
+          * to the pixel block size.
+          */
+         if (info->surf.row_pitch_B * 8 % max_bpb)
+            continue;
+      }
+
+      if (info->view.usage & (ISL_SURF_USAGE_RENDER_TARGET_BIT |
+                              ISL_SURF_USAGE_STORAGE_BIT)) {
+         /* Some destinations require the base address be aligned to the pixel
+          * block size.
+          */
+         if (info->addr.offset * 8 % max_bpb)
+            continue;
+      }
+
+      struct isl_tile_info tile_info;
+      isl_tiling_get_info(info->surf.tiling, info->surf.dim,
+                          info->surf.msaa_layout, max_bpb, info->surf.samples,
+                          &tile_info);
+      assert(surf_tile_info.swiz_count == tile_info.swiz_count);
+      if (memcmp(surf_tile_info.swiz, tile_info.swiz, tile_info.swiz_count))
+         continue;
+
+      if (info->surf.miptail_start_level < info->view.base_level &&
+          surf_tile_info.max_miptail_levels != tile_info.max_miptail_levels)
+         continue;
+
+      return max_bpb / surf_fmtl->bpb;
+   }
+
+   UNREACHABLE("Invalid loop condition above");
+}
+
+static void
+format_scale_copy(const struct isl_device *isl_dev,
+                  struct blorp_surface_info *info,
+                  uint32_t *x, uint32_t *width, int scale)
+{
+   uint32_t orig_fmt_bpb = isl_format_get_layout(info->surf.format)->bpb;
+   info->view.format = get_copy_format_for_bpb(isl_dev, orig_fmt_bpb * scale);
+
+   if (isl_tiling_is_64(info->surf.tiling) ||
+       isl_tiling_is_std_y(info->surf.tiling)) {
+      blorp_surf_convert_to_single_level_tile(isl_dev, info, true);
+   } else {
+      blorp_surf_convert_to_single_slice(isl_dev, info);
+
+      assert(info->surf.logical_level0_px.w == info->surf.phys_level0_sa.w);
+      info->surf.logical_level0_px.w = info->surf.phys_level0_sa.w =
+         DIV_ROUND_UP(info->surf.logical_level0_px.w, scale);
+      info->tile_x_sa /= scale;
+      info->surf.format = info->view.format;
+   }
+
+   *x /= scale;
+   *width = DIV_ROUND_UP(*width, scale);
+}
 
 void
 blorp_copy(struct blorp_batch *batch,
@@ -3167,11 +3318,37 @@ blorp_copy(struct blorp_batch *batch,
       key.need_dst_offset = true;
    }
 
-   /* Once both surfaces are stompped to uncompressed as needed, the
+   /* Once both surfaces are stomped to uncompressed as needed, the
-    * destination size is the same as the source size.
+    * destination size is the same as the source size unless we're copying
+    * between YUV and color images. We'll remove any differences in the
+    * process of using the largest format possible for the copy.
     */
    uint32_t dst_width = src_width;
    uint32_t dst_height = src_height;
+   if (isl_format_is_yuv(src_fmtl->format) !=
+       isl_format_is_yuv(dst_fmtl->format))
+      dst_width *= src_fmtl->bpb / dst_fmtl->bpb;
+
+   int max_fmt_scale_src = get_max_format_scale(isl_dev, &params.src, src_x,
+                                                src_width, src_height);
+   int max_fmt_scale_dst = get_max_format_scale(isl_dev, &params.dst, dst_x,
+                                                dst_width, dst_height);
+   int copy_fmt_bpb = MIN2(src_fmtl->bpb * max_fmt_scale_src,
+                           dst_fmtl->bpb * max_fmt_scale_dst);
+
+   if (src_fmtl->bpb < copy_fmt_bpb) {
+      format_scale_copy(isl_dev, &params.src, &src_x, &src_width,
+                        copy_fmt_bpb / src_fmtl->bpb);
+      key.need_src_offset = true;
+   }
+
+   if (dst_fmtl->bpb < copy_fmt_bpb) {
+      format_scale_copy(isl_dev, &params.dst, &dst_x, &dst_width,
+                        copy_fmt_bpb / dst_fmtl->bpb);
+      key.need_dst_offset = true;
+   }
+
+   assert(src_width == dst_width);
 
    if (params.src.view.format != params.dst.view.format) {
       enum isl_format src_cast_format = params.src.view.format;