radeonsi: add new SDMA texture copy code

This implements:
- Linear-to-linear partial copies. (unaligned)
- Tiled-to-linear and linear-to-tiled partial copies.
  (unaligned except 1-2 Bpp)
- Tiled-to-tiled partial copies aligned to 8x8.

v2: Extend the SDMA L2T VM fault workaround to T2L.
    - Same algorithm, just applied to T2L.
      (and using a 0-based address and surface.bo_size instead of buf->size)

Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

1 files changed, 438 insertions, 0 deletions

diff --git a/src/gallium/drivers/radeonsi/cik_sdma.c b/src/gallium/drivers/radeonsi/cik_sdma.c
index 5021578871b..396cecc4b44 100644
--- a/src/gallium/drivers/radeonsi/cik_sdma.c
+++ b/src/gallium/drivers/radeonsi/cik_sdma.c
@@ -90,6 +90,440 @@ static void cik_sdma_copy_buffer(struct si_context *ctx,
 	r600_dma_emit_wait_idle(&ctx->b);
 }
 
+static unsigned minify_as_blocks(unsigned width, unsigned level, unsigned blk_w)
+{
+	width = u_minify(width, level);
+	return DIV_ROUND_UP(width, blk_w);
+}
+
+static unsigned encode_tile_info(struct si_context *sctx,
+				 struct r600_texture *tex, unsigned level,
+				 bool set_bpp)
+{
+	struct radeon_info *info = &sctx->screen->b.info;
+	unsigned tile_index = tex->surface.tiling_index[level];
+	unsigned macro_tile_index = tex->surface.macro_tile_index;
+	unsigned tile_mode = info->si_tile_mode_array[tile_index];
+	unsigned macro_tile_mode = info->cik_macrotile_mode_array[macro_tile_index];
+
+	return (set_bpp ? util_logbase2(tex->surface.bpe) : 0) |
+		(G_009910_ARRAY_MODE(tile_mode) << 3) |
+		(G_009910_MICRO_TILE_MODE_NEW(tile_mode) << 8) |
+		/* Non-depth modes don't have TILE_SPLIT set. */
+		((util_logbase2(tex->surface.tile_split >> 6)) << 11) |
+		(G_009990_BANK_WIDTH(macro_tile_mode) << 15) |
+		(G_009990_BANK_HEIGHT(macro_tile_mode) << 18) |
+		(G_009990_NUM_BANKS(macro_tile_mode) << 21) |
+		(G_009990_MACRO_TILE_ASPECT(macro_tile_mode) << 24) |
+		(G_009910_PIPE_CONFIG(tile_mode) << 26);
+}
+
+static bool cik_sdma_copy_texture(struct si_context *sctx,
+				  struct pipe_resource *dst,
+				  unsigned dst_level,
+				  unsigned dstx, unsigned dsty, unsigned dstz,
+				  struct pipe_resource *src,
+				  unsigned src_level,
+				  const struct pipe_box *src_box)
+{
+	struct radeon_info *info = &sctx->screen->b.info;
+	struct r600_texture *rsrc = (struct r600_texture*)src;
+	struct r600_texture *rdst = (struct r600_texture*)dst;
+	unsigned bpp = rdst->surface.bpe;
+	uint64_t dst_address = rdst->resource.gpu_address +
+			       rdst->surface.level[dst_level].offset;
+	uint64_t src_address = rsrc->resource.gpu_address +
+			       rsrc->surface.level[src_level].offset;
+	unsigned dst_mode = rdst->surface.level[dst_level].mode;
+	unsigned src_mode = rsrc->surface.level[src_level].mode;
+	unsigned dst_tile_index = rdst->surface.tiling_index[dst_level];
+	unsigned src_tile_index = rsrc->surface.tiling_index[src_level];
+	unsigned dst_tile_mode = info->si_tile_mode_array[dst_tile_index];
+	unsigned src_tile_mode = info->si_tile_mode_array[src_tile_index];
+	unsigned dst_micro_mode = G_009910_MICRO_TILE_MODE_NEW(dst_tile_mode);
+	unsigned src_micro_mode = G_009910_MICRO_TILE_MODE_NEW(src_tile_mode);
+	unsigned dst_pitch = rdst->surface.level[dst_level].pitch_bytes / bpp;
+	unsigned src_pitch = rsrc->surface.level[src_level].pitch_bytes / bpp;
+	uint64_t dst_slice_pitch = rdst->surface.level[dst_level].slice_size / bpp;
+	uint64_t src_slice_pitch = rsrc->surface.level[src_level].slice_size / bpp;
+	unsigned dst_width = minify_as_blocks(rdst->resource.b.b.width0,
+					      dst_level, rdst->surface.blk_w);
+	unsigned src_width = minify_as_blocks(rsrc->resource.b.b.width0,
+					      src_level, rsrc->surface.blk_w);
+	unsigned dst_height = minify_as_blocks(rdst->resource.b.b.height0,
+					       dst_level, rdst->surface.blk_h);
+	unsigned src_height = minify_as_blocks(rsrc->resource.b.b.height0,
+					       src_level, rsrc->surface.blk_h);
+	unsigned srcx = src_box->x / rsrc->surface.blk_w;
+	unsigned srcy = src_box->y / rsrc->surface.blk_h;
+	unsigned srcz = src_box->z;
+	unsigned copy_width = DIV_ROUND_UP(src_box->width, rsrc->surface.blk_w);
+	unsigned copy_height = DIV_ROUND_UP(src_box->height, rsrc->surface.blk_h);
+	unsigned copy_depth = src_box->depth;
+
+	assert(src_level <= src->last_level);
+	assert(dst_level <= dst->last_level);
+	assert(rdst->surface.level[dst_level].offset +
+	       dst_slice_pitch * bpp * (dstz + src_box->depth) <=
+	       rdst->resource.buf->size);
+	assert(rsrc->surface.level[src_level].offset +
+	       src_slice_pitch * bpp * (srcz + src_box->depth) <=
+	       rsrc->resource.buf->size);
+
+	/* Test CIK with radeon and amdgpu before enabling this. */
+	if (sctx->b.chip_class == CIK)
+		return false;
+
+	if (!r600_prepare_for_dma_blit(&sctx->b, rdst, dst_level, dstx, dsty,
+					dstz, rsrc, src_level, src_box))
+		return false;
+
+	dstx /= rdst->surface.blk_w;
+	dsty /= rdst->surface.blk_h;
+
+	if (srcx >= (1 << 14) ||
+	    srcy >= (1 << 14) ||
+	    srcz >= (1 << 11) ||
+	    dstx >= (1 << 14) ||
+	    dsty >= (1 << 14) ||
+	    dstz >= (1 << 11))
+		return false;
+
+	/* Linear -> linear sub-window copy. */
+	if (dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED &&
+	    src_mode == RADEON_SURF_MODE_LINEAR_ALIGNED &&
+	    /* check if everything fits into the bitfields */
+	    src_pitch <= (1 << 14) &&
+	    dst_pitch <= (1 << 14) &&
+	    src_slice_pitch <= (1 << 28) &&
+	    dst_slice_pitch <= (1 << 28) &&
+	    copy_width <= (1 << 14) &&
+	    copy_height <= (1 << 14) &&
+	    copy_depth <= (1 << 11) &&
+	    /* HW limitation - CIK: */
+	    (sctx->b.chip_class != CIK ||
+	     (copy_width < (1 << 14) &&
+	      copy_height < (1 << 14) &&
+	      copy_depth < (1 << 11))) &&
+	    /* HW limitation - some CIK parts: */
+	    ((sctx->b.family != CHIP_BONAIRE &&
+	      sctx->b.family != CHIP_KAVERI) ||
+	     (srcx + copy_width != (1 << 14) &&
+	      srcy + copy_height != (1 << 14)))) {
+		struct radeon_winsys_cs *cs = sctx->b.dma.cs;
+
+		r600_need_dma_space(&sctx->b, 13);
+		radeon_add_to_buffer_list(&sctx->b, &sctx->b.dma, &rsrc->resource,
+					  RADEON_USAGE_READ,
+					  RADEON_PRIO_SDMA_TEXTURE);
+		radeon_add_to_buffer_list(&sctx->b, &sctx->b.dma, &rdst->resource,
+					  RADEON_USAGE_WRITE,
+					  RADEON_PRIO_SDMA_TEXTURE);
+
+		radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY,
+						CIK_SDMA_COPY_SUB_OPCODE_LINEAR_SUB_WINDOW, 0) |
+			    (util_logbase2(bpp) << 29));
+		radeon_emit(cs, src_address);
+		radeon_emit(cs, src_address >> 32);
+		radeon_emit(cs, srcx | (srcy << 16));
+		radeon_emit(cs, srcz | ((src_pitch - 1) << 16));
+		radeon_emit(cs, src_slice_pitch - 1);
+		radeon_emit(cs, dst_address);
+		radeon_emit(cs, dst_address >> 32);
+		radeon_emit(cs, dstx | (dsty << 16));
+		radeon_emit(cs, dstz | ((dst_pitch - 1) << 16));
+		radeon_emit(cs, dst_slice_pitch - 1);
+		if (sctx->b.chip_class == CIK) {
+			radeon_emit(cs, copy_width | (copy_height << 16));
+			radeon_emit(cs, copy_depth);
+		} else {
+			radeon_emit(cs, (copy_width - 1) | ((copy_height - 1) << 16));
+			radeon_emit(cs, (copy_depth - 1));
+		}
+
+		r600_dma_emit_wait_idle(&sctx->b);
+		return true;
+	}
+
+	/* Tiled <-> linear sub-window copy. */
+	if ((src_mode >= RADEON_SURF_MODE_1D) != (dst_mode >= RADEON_SURF_MODE_1D)) {
+		struct r600_texture *tiled = src_mode >= RADEON_SURF_MODE_1D ? rsrc : rdst;
+		struct r600_texture *linear = tiled == rsrc ? rdst : rsrc;
+		unsigned tiled_level =	tiled	== rsrc ? src_level : dst_level;
+		unsigned linear_level =	linear	== rsrc ? src_level : dst_level;
+		unsigned tiled_x =	tiled	== rsrc ? srcx : dstx;
+		unsigned linear_x =	linear  == rsrc ? srcx : dstx;
+		unsigned tiled_y =	tiled	== rsrc ? srcy : dsty;
+		unsigned linear_y =	linear  == rsrc ? srcy : dsty;
+		unsigned tiled_z =	tiled	== rsrc ? srcz : dstz;
+		unsigned linear_z =	linear  == rsrc ? srcz : dstz;
+		unsigned tiled_width =	tiled	== rsrc ? src_width : dst_width;
+		unsigned linear_width =	linear	== rsrc ? src_width : dst_width;
+		unsigned tiled_pitch =	tiled	== rsrc ? src_pitch : dst_pitch;
+		unsigned linear_pitch =	linear	== rsrc ? src_pitch : dst_pitch;
+		unsigned tiled_slice_pitch  = tiled  == rsrc ? src_slice_pitch : dst_slice_pitch;
+		unsigned linear_slice_pitch = linear == rsrc ? src_slice_pitch : dst_slice_pitch;
+		uint64_t tiled_address =  tiled  == rsrc ? src_address : dst_address;
+		uint64_t linear_address = linear == rsrc ? src_address : dst_address;
+		unsigned tiled_micro_mode = tiled == rsrc ? src_micro_mode : dst_micro_mode;
+
+		assert(tiled_pitch % 8 == 0);
+		assert(tiled_slice_pitch % 64 == 0);
+		unsigned pitch_tile_max = tiled_pitch / 8 - 1;
+		unsigned slice_tile_max = tiled_slice_pitch / 64 - 1;
+		unsigned xalign = MAX2(1, 4 / bpp);
+		unsigned copy_width_aligned = copy_width;
+
+		/* If the region ends at the last pixel and is unaligned, we
+		 * can copy the remainder of the line that is not visible to
+		 * make it aligned.
+		 */
+		if (copy_width % xalign != 0 &&
+		    linear_x + copy_width == linear_width &&
+		    tiled_x  + copy_width == tiled_width &&
+		    linear_x + align(copy_width, xalign) <= linear_pitch &&
+		    tiled_x  + align(copy_width, xalign) <= tiled_pitch)
+			copy_width_aligned = align(copy_width, xalign);
+
+		/* HW limitations. */
+		if ((sctx->b.family == CHIP_BONAIRE ||
+		     sctx->b.family == CHIP_KAVERI) &&
+		    linear_pitch - 1 == 0x3fff &&
+		    bpp == 16)
+			return false;
+
+		if (sctx->b.chip_class == CIK &&
+		    (copy_width_aligned == (1 << 14) ||
+		     copy_height == (1 << 14) ||
+		     copy_depth == (1 << 11)))
+			return false;
+
+		if ((sctx->b.family == CHIP_BONAIRE ||
+		     sctx->b.family == CHIP_KAVERI ||
+		     sctx->b.family == CHIP_KABINI ||
+		     sctx->b.family == CHIP_MULLINS) &&
+		    (tiled_x + copy_width == (1 << 14) ||
+		     tiled_y + copy_height == (1 << 14)))
+			return false;
+
+		/* The hw can read outside of the given linear buffer bounds,
+		 * or access those pages but not touch the memory in case
+		 * of writes. (it still causes a VM fault)
+		 *
+		 * Out-of-bounds memory access or page directory access must
+		 * be prevented.
+		 */
+		int64_t start_linear_address, end_linear_address;
+		unsigned granularity;
+
+		/* Deduce the size of reads from the linear surface. */
+		switch (tiled_micro_mode) {
+		case V_009910_ADDR_SURF_DISPLAY_MICRO_TILING:
+			granularity = bpp == 1 ? 64 / (8*bpp) :
+						 128 / (8*bpp);
+			break;
+		case V_009910_ADDR_SURF_THIN_MICRO_TILING:
+		case V_009910_ADDR_SURF_DEPTH_MICRO_TILING:
+			if (0 /* TODO: THICK microtiling */)
+				granularity = bpp == 1 ? 32 / (8*bpp) :
+					      bpp == 2 ? 64 / (8*bpp) :
+					      bpp <= 8 ? 128 / (8*bpp) :
+							 256 / (8*bpp);
+			else
+				granularity = bpp <= 2 ? 64 / (8*bpp) :
+					      bpp <= 8 ? 128 / (8*bpp) :
+							 256 / (8*bpp);
+			break;
+		default:
+			return false;
+		}
+
+		/* The linear reads start at tiled_x & ~(granularity - 1).
+		 * If linear_x == 0 && tiled_x % granularity != 0, the hw
+		 * starts reading from an address preceding linear_address!!!
+		 */
+		start_linear_address =
+			linear->surface.level[linear_level].offset +
+			bpp * (linear_z * linear_slice_pitch +
+			       linear_y * linear_pitch +
+			       linear_x);
+		start_linear_address -= (int)(bpp * (tiled_x % granularity));
+
+		end_linear_address =
+			linear->surface.level[linear_level].offset +
+			bpp * ((linear_z + copy_depth - 1) * linear_slice_pitch +
+			       (linear_y + copy_height - 1) * linear_pitch +
+			       (linear_x + copy_width));
+
+		if ((tiled_x + copy_width) % granularity)
+			end_linear_address += granularity -
+					      (tiled_x + copy_width) % granularity;
+
+		if (start_linear_address < 0 ||
+		    end_linear_address > linear->surface.bo_size)
+			return false;
+
+		/* Check requirements. */
+		if (tiled_address % 256 == 0 &&
+		    linear_address % 4 == 0 &&
+		    linear_pitch % xalign == 0 &&
+		    linear_x % xalign == 0 &&
+		    tiled_x % xalign == 0 &&
+		    copy_width_aligned % xalign == 0 &&
+		    tiled_micro_mode != V_009910_ADDR_SURF_ROTATED_MICRO_TILING &&
+		    /* check if everything fits into the bitfields */
+		    tiled->surface.tile_split <= 4096 &&
+		    pitch_tile_max < (1 << 11) &&
+		    slice_tile_max < (1 << 22) &&
+		    linear_pitch <= (1 << 14) &&
+		    linear_slice_pitch <= (1 << 28) &&
+		    copy_width_aligned <= (1 << 14) &&
+		    copy_height <= (1 << 14) &&
+		    copy_depth <= (1 << 11)) {
+			struct radeon_winsys_cs *cs = sctx->b.dma.cs;
+
+			r600_need_dma_space(&sctx->b, 14);
+			radeon_add_to_buffer_list(&sctx->b, &sctx->b.dma, &rsrc->resource,
+						  RADEON_USAGE_READ,
+						  RADEON_PRIO_SDMA_TEXTURE);
+			radeon_add_to_buffer_list(&sctx->b, &sctx->b.dma, &rdst->resource,
+						  RADEON_USAGE_WRITE,
+						  RADEON_PRIO_SDMA_TEXTURE);
+
+			radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY,
+							CIK_SDMA_COPY_SUB_OPCODE_TILED_SUB_WINDOW, 0) |
+					((linear == rdst) << 31));
+			radeon_emit(cs, tiled_address);
+			radeon_emit(cs, tiled_address >> 32);
+			radeon_emit(cs, tiled_x | (tiled_y << 16));
+			radeon_emit(cs, tiled_z | (pitch_tile_max << 16));
+			radeon_emit(cs, slice_tile_max);
+			radeon_emit(cs, encode_tile_info(sctx, tiled, tiled_level, true));
+			radeon_emit(cs, linear_address);
+			radeon_emit(cs, linear_address >> 32);
+			radeon_emit(cs, linear_x | (linear_y << 16));
+			radeon_emit(cs, linear_z | ((linear_pitch - 1) << 16));
+			radeon_emit(cs, linear_slice_pitch - 1);
+			if (sctx->b.chip_class == CIK) {
+				radeon_emit(cs, copy_width_aligned | (copy_height << 16));
+				radeon_emit(cs, copy_depth);
+			} else {
+				radeon_emit(cs, (copy_width_aligned - 1) | ((copy_height - 1) << 16));
+				radeon_emit(cs, (copy_depth - 1));
+			}
+
+			r600_dma_emit_wait_idle(&sctx->b);
+			return true;
+		}
+	}
+
+	/* Tiled -> Tiled sub-window copy. */
+	if (dst_mode >= RADEON_SURF_MODE_1D &&
+	    src_mode >= RADEON_SURF_MODE_1D &&
+	    /* check if these fit into the bitfields */
+	    src_address % 256 == 0 &&
+	    dst_address % 256 == 0 &&
+	    rsrc->surface.tile_split <= 4096 &&
+	    rdst->surface.tile_split <= 4096 &&
+	    dstx % 8 == 0 &&
+	    dsty % 8 == 0 &&
+	    srcx % 8 == 0 &&
+	    srcy % 8 == 0 &&
+	    /* this can either be equal, or display->rotated (VI only) */
+	    (src_micro_mode == dst_micro_mode ||
+	     (sctx->b.chip_class == VI &&
+	      src_micro_mode == V_009910_ADDR_SURF_DISPLAY_MICRO_TILING &&
+	      dst_micro_mode == V_009910_ADDR_SURF_ROTATED_MICRO_TILING))) {
+		assert(src_pitch % 8 == 0);
+		assert(dst_pitch % 8 == 0);
+		assert(src_slice_pitch % 64 == 0);
+		assert(dst_slice_pitch % 64 == 0);
+		unsigned src_pitch_tile_max = src_pitch / 8 - 1;
+		unsigned dst_pitch_tile_max = dst_pitch / 8 - 1;
+		unsigned src_slice_tile_max = src_slice_pitch / 64 - 1;
+		unsigned dst_slice_tile_max = dst_slice_pitch / 64 - 1;
+		unsigned copy_width_aligned = copy_width;
+		unsigned copy_height_aligned = copy_height;
+
+		/* If the region ends at the last pixel and is unaligned, we
+		 * can copy the remainder of the tile that is not visible to
+		 * make it aligned.
+		 */
+		if (copy_width % 8 != 0 &&
+		    srcx + copy_width == src_width &&
+		    dstx + copy_width == dst_width)
+			copy_width_aligned = align(copy_width, 8);
+
+		if (copy_height % 8 != 0 &&
+		    srcy + copy_height == src_height &&
+		    dsty + copy_height == dst_height)
+			copy_height_aligned = align(copy_height, 8);
+
+		/* check if these fit into the bitfields */
+		if (src_pitch_tile_max < (1 << 11) &&
+		    dst_pitch_tile_max < (1 << 11) &&
+		    src_slice_tile_max < (1 << 22) &&
+		    dst_slice_tile_max < (1 << 22) &&
+		    copy_width_aligned <= (1 << 14) &&
+		    copy_height_aligned <= (1 << 14) &&
+		    copy_depth <= (1 << 11) &&
+		    copy_width_aligned % 8 == 0 &&
+		    copy_height_aligned % 8 == 0 &&
+		    /* HW limitation - CIK: */
+		    (sctx->b.chip_class != CIK ||
+		     (copy_width_aligned < (1 << 14) &&
+		      copy_height_aligned < (1 << 14) &&
+		      copy_depth < (1 << 11))) &&
+		    /* HW limitation - some CIK parts: */
+		    ((sctx->b.family != CHIP_BONAIRE &&
+		      sctx->b.family != CHIP_KAVERI &&
+		      sctx->b.family != CHIP_KABINI &&
+		      sctx->b.family != CHIP_MULLINS) ||
+		     (srcx + copy_width_aligned != (1 << 14) &&
+		      srcy + copy_height_aligned != (1 << 14) &&
+		      dstx + copy_width != (1 << 14)))) {
+			struct radeon_winsys_cs *cs = sctx->b.dma.cs;
+
+			r600_need_dma_space(&sctx->b, 15);
+			radeon_add_to_buffer_list(&sctx->b, &sctx->b.dma, &rsrc->resource,
+						  RADEON_USAGE_READ,
+						  RADEON_PRIO_SDMA_TEXTURE);
+			radeon_add_to_buffer_list(&sctx->b, &sctx->b.dma, &rdst->resource,
+						  RADEON_USAGE_WRITE,
+						  RADEON_PRIO_SDMA_TEXTURE);
+
+			radeon_emit(cs, CIK_SDMA_PACKET(CIK_SDMA_OPCODE_COPY,
+							CIK_SDMA_COPY_SUB_OPCODE_T2T_SUB_WINDOW, 0));
+			radeon_emit(cs, src_address);
+			radeon_emit(cs, src_address >> 32);
+			radeon_emit(cs, srcx | (srcy << 16));
+			radeon_emit(cs, srcz | (src_pitch_tile_max << 16));
+			radeon_emit(cs, src_slice_tile_max);
+			radeon_emit(cs, encode_tile_info(sctx, rsrc, src_level, true));
+			radeon_emit(cs, dst_address);
+			radeon_emit(cs, dst_address >> 32);
+			radeon_emit(cs, dstx | (dsty << 16));
+			radeon_emit(cs, dstz | (dst_pitch_tile_max << 16));
+			radeon_emit(cs, dst_slice_tile_max);
+			radeon_emit(cs, encode_tile_info(sctx, rdst, dst_level, false));
+			if (sctx->b.chip_class == CIK) {
+				radeon_emit(cs, copy_width_aligned |
+						(copy_height_aligned << 16));
+				radeon_emit(cs, copy_depth);
+			} else {
+				radeon_emit(cs, (copy_width_aligned - 8) |
+						((copy_height_aligned - 8) << 16));
+				radeon_emit(cs, (copy_depth - 1));
+			}
+
+			r600_dma_emit_wait_idle(&sctx->b);
+			return true;
+		}
+	}
+
+	return false;
+}
+
 static void cik_sdma_copy(struct pipe_context *ctx,
 			  struct pipe_resource *dst,
 			  unsigned dst_level,
@@ -108,6 +542,10 @@ static void cik_sdma_copy(struct pipe_context *ctx,
 		return;
 	}
 
+	if (cik_sdma_copy_texture(sctx, dst, dst_level, dstx, dsty, dstz,
+				  src, src_level, src_box))
+		return;
+
 fallback:
 	si_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
 				src, src_level, src_box);