radeonsi: move HTILE allocation outside of radeonsi

ac_surface computes it for amdgpu.
radeon_drm_surface computes it for radeon.

Acked-by: Pierre-Eric Pelloux-Prayer <[email protected]>

4 files changed, 93 insertions, 91 deletions

diff --git a/src/amd/common/ac_surface.c b/src/amd/common/ac_surface.c
index 2c2917d4a23..1d254ec3a78 100644
--- a/src/amd/common/ac_surface.c
+++ b/src/amd/common/ac_surface.c
@@ -338,11 +338,12 @@ static int gfx6_compute_level(ADDR_HANDLE addrlib,
 		}
 	}
 
-	/* TC-compatible HTILE. */
+	/* HTILE. */
 	if (!is_stencil &&
 	    AddrSurfInfoIn->flags.depth &&
 	    surf_level->mode == RADEON_SURF_MODE_2D &&
-	    level == 0) {
+	    level == 0 &&
+	    !(surf->flags & RADEON_SURF_NO_HTILE)) {
 		AddrHtileIn->flags.tcCompatible = AddrSurfInfoIn->flags.tcCompatible;
 		AddrHtileIn->pitch = AddrSurfInfoOut->pitch;
 		AddrHtileIn->height = AddrSurfInfoOut->height;
@@ -1065,6 +1066,9 @@ static int gfx9_compute_miptree(ADDR_HANDLE addrlib,
 	if (in->flags.depth) {
 		assert(in->swizzleMode != ADDR_SW_LINEAR);
 
+		if (surf->flags & RADEON_SURF_NO_HTILE)
+			return 0;
+
 		/* HTILE */
 		ADDR2_COMPUTE_HTILE_INFO_INPUT hin = {0};
 		ADDR2_COMPUTE_HTILE_INFO_OUTPUT hout = {0};
@@ -1091,7 +1095,10 @@ static int gfx9_compute_miptree(ADDR_HANDLE addrlib,
 		surf->htile_size = hout.htileBytes;
 		surf->htile_slice_size = hout.sliceSize;
 		surf->htile_alignment = hout.baseAlign;
-	} else {
+		return 0;
+	}
+
+	{
 		/* Compute tile swizzle for the color surface.
 		 * All *_X and *_T modes can use the swizzle.
 		 */
diff --git a/src/amd/common/ac_surface.h b/src/amd/common/ac_surface.h
index 53074c90faf..52aa63bff2e 100644
--- a/src/amd/common/ac_surface.h
+++ b/src/amd/common/ac_surface.h
@@ -72,6 +72,7 @@ enum radeon_micro_mode {
 #define RADEON_SURF_NO_RENDER_TARGET            (1 << 27)
 #define RADEON_SURF_FORCE_SWIZZLE_MODE          (1 << 28)
 #define RADEON_SURF_NO_FMASK                    (1 << 29)
+#define RADEON_SURF_NO_HTILE                    (1 << 30)
 
 struct legacy_surf_level {
     uint64_t                    offset;
diff --git a/src/gallium/drivers/radeonsi/si_texture.c b/src/gallium/drivers/radeonsi/si_texture.c
index baa34d0b408..439de3d0ffa 100644
--- a/src/gallium/drivers/radeonsi/si_texture.c
+++ b/src/gallium/drivers/radeonsi/si_texture.c
@@ -253,9 +253,11 @@ static int si_init_surface(struct si_screen *sscreen,
 	if (!is_flushed_depth && is_depth) {
 		flags |= RADEON_SURF_ZBUFFER;
 
-		if (tc_compatible_htile &&
-		    (sscreen->info.chip_class >= GFX9 ||
-		     array_mode == RADEON_SURF_MODE_2D)) {
+		if (sscreen->debug_flags & DBG(NO_HYPERZ)) {
+			flags |= RADEON_SURF_NO_HTILE;
+		} else if (tc_compatible_htile &&
+			   (sscreen->info.chip_class >= GFX9 ||
+			    array_mode == RADEON_SURF_MODE_2D)) {
 			/* TC-compatible HTILE only supports Z32_FLOAT.
 			 * GFX9 also supports Z16_UNORM.
 			 * On GFX8, promote Z16 to Z32. DB->CB copies will convert
@@ -1067,85 +1069,6 @@ static void si_texture_destroy(struct pipe_screen *screen,
 
 static const struct u_resource_vtbl si_texture_vtbl;
 
-static void si_texture_get_htile_size(struct si_screen *sscreen,
-				      struct si_texture *tex)
-{
-	unsigned cl_width, cl_height, width, height;
-	unsigned slice_elements, slice_bytes, pipe_interleave_bytes, base_align;
-	unsigned num_pipes = sscreen->info.num_tile_pipes;
-
-	assert(sscreen->info.chip_class <= GFX8);
-
-	tex->surface.htile_size = 0;
-
-	if (tex->surface.u.legacy.level[0].mode == RADEON_SURF_MODE_1D &&
-	    !sscreen->info.htile_cmask_support_1d_tiling)
-		return;
-
-	/* Overalign HTILE on P2 configs to work around GPU hangs in
-	 * piglit/depthstencil-render-miplevels 585.
-	 *
-	 * This has been confirmed to help Kabini & Stoney, where the hangs
-	 * are always reproducible. I think I have seen the test hang
-	 * on Carrizo too, though it was very rare there.
-	 */
-	if (sscreen->info.chip_class >= GFX7 && num_pipes < 4)
-		num_pipes = 4;
-
-	switch (num_pipes) {
-	case 1:
-		cl_width = 32;
-		cl_height = 16;
-		break;
-	case 2:
-		cl_width = 32;
-		cl_height = 32;
-		break;
-	case 4:
-		cl_width = 64;
-		cl_height = 32;
-		break;
-	case 8:
-		cl_width = 64;
-		cl_height = 64;
-		break;
-	case 16:
-		cl_width = 128;
-		cl_height = 64;
-		break;
-	default:
-		assert(0);
-		return;
-	}
-
-	width = align(tex->surface.u.legacy.level[0].nblk_x, cl_width * 8);
-	height = align(tex->surface.u.legacy.level[0].nblk_y, cl_height * 8);
-
-	slice_elements = (width * height) / (8 * 8);
-	slice_bytes = slice_elements * 4;
-
-	pipe_interleave_bytes = sscreen->info.pipe_interleave_bytes;
-	base_align = num_pipes * pipe_interleave_bytes;
-
-	tex->surface.htile_alignment = base_align;
-	tex->surface.htile_size =
-		util_num_layers(&tex->buffer.b.b, 0) *
-		align(slice_bytes, base_align);
-}
-
-static void si_texture_allocate_htile(struct si_screen *sscreen,
-				      struct si_texture *tex)
-{
-	if (sscreen->info.chip_class <= GFX8 && !tex->tc_compatible_htile)
-		si_texture_get_htile_size(sscreen, tex);
-
-	if (!tex->surface.htile_size)
-		return;
-
-	tex->htile_offset = align(tex->size, tex->surface.htile_alignment);
-	tex->size = tex->htile_offset + tex->surface.htile_size;
-}
-
 void si_print_texture_info(struct si_screen *sscreen,
 			   struct si_texture *tex, struct u_log_context *log)
 {
@@ -1370,12 +1293,12 @@ si_texture_create_object(struct pipe_screen *screen,
 			tex->can_sample_s = !tex->surface.u.legacy.stencil_adjusted;
 		}
 
-		if (!(base->flags & (SI_RESOURCE_FLAG_TRANSFER |
-				     SI_RESOURCE_FLAG_FLUSHED_DEPTH))) {
-			tex->db_compatible = true;
+		tex->db_compatible = surface->flags & RADEON_SURF_ZBUFFER;
 
-			if (!(sscreen->debug_flags & DBG(NO_HYPERZ)))
-				si_texture_allocate_htile(sscreen, tex);
+		if (tex->surface.htile_size) {
+			tex->htile_offset = align64(tex->size,
+						    tex->surface.htile_alignment);
+			tex->size = tex->htile_offset + tex->surface.htile_size;
 		}
 	} else {
 		if (tex->surface.fmask_size) {
@@ -1678,7 +1601,8 @@ struct pipe_resource *si_texture_create(struct pipe_screen *screen,
 	}
 
 	struct radeon_surf surface = {0};
-	bool is_flushed_depth = templ->flags & SI_RESOURCE_FLAG_FLUSHED_DEPTH;
+	bool is_flushed_depth = templ->flags & SI_RESOURCE_FLAG_FLUSHED_DEPTH ||
+				templ->flags & SI_RESOURCE_FLAG_TRANSFER;
 	bool tc_compatible_htile =
 		sscreen->info.chip_class >= GFX8 &&
 		/* There are issues with TC-compatible HTILE on Tonga (and
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_surface.c b/src/gallium/winsys/radeon/drm/radeon_drm_surface.c
index e09805784e0..332b9e4a428 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_surface.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_surface.c
@@ -280,6 +280,72 @@ static void si_compute_cmask(const struct radeon_info *info,
 	surf->cmask_size = align(slice_bytes, base_align) * num_layers;
 }
 
+static void si_compute_htile(const struct radeon_info *info,
+                             struct radeon_surf *surf, unsigned num_layers)
+{
+    unsigned cl_width, cl_height, width, height;
+    unsigned slice_elements, slice_bytes, pipe_interleave_bytes, base_align;
+    unsigned num_pipes = info->num_tile_pipes;
+
+    surf->htile_size = 0;
+
+    if (!(surf->flags & RADEON_SURF_Z_OR_SBUFFER) ||
+        surf->flags & RADEON_SURF_NO_HTILE)
+        return;
+
+    if (surf->u.legacy.level[0].mode == RADEON_SURF_MODE_1D &&
+        !info->htile_cmask_support_1d_tiling)
+        return;
+
+    /* Overalign HTILE on P2 configs to work around GPU hangs in
+     * piglit/depthstencil-render-miplevels 585.
+     *
+     * This has been confirmed to help Kabini & Stoney, where the hangs
+     * are always reproducible. I think I have seen the test hang
+     * on Carrizo too, though it was very rare there.
+     */
+    if (info->chip_class >= GFX7 && num_pipes < 4)
+        num_pipes = 4;
+
+    switch (num_pipes) {
+    case 1:
+        cl_width = 32;
+        cl_height = 16;
+        break;
+    case 2:
+        cl_width = 32;
+        cl_height = 32;
+        break;
+    case 4:
+        cl_width = 64;
+        cl_height = 32;
+        break;
+    case 8:
+        cl_width = 64;
+        cl_height = 64;
+        break;
+    case 16:
+        cl_width = 128;
+        cl_height = 64;
+        break;
+    default:
+        assert(0);
+        return;
+    }
+
+    width = align(surf->u.legacy.level[0].nblk_x, cl_width * 8);
+    height = align(surf->u.legacy.level[0].nblk_y, cl_height * 8);
+
+    slice_elements = (width * height) / (8 * 8);
+    slice_bytes = slice_elements * 4;
+
+    pipe_interleave_bytes = info->pipe_interleave_bytes;
+    base_align = num_pipes * pipe_interleave_bytes;
+
+    surf->htile_alignment = base_align;
+    surf->htile_size = num_layers * align(slice_bytes, base_align);
+}
+
 static int radeon_winsys_surface_init(struct radeon_winsys *rws,
                                       const struct pipe_resource *tex,
                                       unsigned flags, unsigned bpe,
@@ -365,6 +431,10 @@ static int radeon_winsys_surface_init(struct radeon_winsys *rws,
 
 	    si_compute_cmask(&ws->info, &config, surf_ws);
     }
+
+    if (ws->gen == DRV_SI)
+        si_compute_htile(&ws->info, surf_ws, util_num_layers(tex, 0));
+
     return 0;
 }