r600g: deobfuscate async DMA code

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

6 files changed, 35 insertions, 31 deletions

diff --git a/src/gallium/drivers/r600/evergreen_hw_context.c b/src/gallium/drivers/r600/evergreen_hw_context.c
index e43eacc9a0a..a47461c20ea 100644
--- a/src/gallium/drivers/r600/evergreen_hw_context.c
+++ b/src/gallium/drivers/r600/evergreen_hw_context.c
@@ -50,19 +50,19 @@ void evergreen_dma_copy(struct r600_context *rctx,
 	src_offset += r600_resource_va(&rctx->screen->b.b, src);
 
 	/* see if we use dword or byte copy */
-	if (!(dst_offset & 0x3) && !(src_offset & 0x3) && !(size & 0x3)) {
+	if (!(dst_offset % 4) && !(src_offset % 4) && !(size % 4)) {
 		size >>= 2;
-		sub_cmd = 0x00;
+		sub_cmd = EG_DMA_COPY_DWORD_ALIGNED;
 		shift = 2;
 	} else {
-		sub_cmd = 0x40;
+		sub_cmd = EG_DMA_COPY_BYTE_ALIGNED;
 		shift = 0;
 	}
-	ncopy = (size / 0x000fffff) + !!(size % 0x000fffff);
+	ncopy = (size / EG_DMA_COPY_MAX_SIZE) + !!(size % EG_DMA_COPY_MAX_SIZE);
 
 	r600_need_dma_space(&rctx->b, ncopy * 5);
 	for (i = 0; i < ncopy; i++) {
-		csize = size < 0x000fffff ? size : 0x000fffff;
+		csize = size < EG_DMA_COPY_MAX_SIZE ? size : EG_DMA_COPY_MAX_SIZE;
 		/* emit reloc before writting cs so that cs is always in consistent state */
 		r600_context_bo_reloc(&rctx->b, &rctx->b.rings.dma, rsrc, RADEON_USAGE_READ,
 				      RADEON_PRIO_MIN);
diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index fb3450674f1..11cd76772af 100644
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -3233,15 +3233,15 @@ static void evergreen_dma_copy_tile(struct r600_context *rctx,
 		non_disp_tiling = 1;
 
 	y = 0;
-	sub_cmd = 0x8;
+	sub_cmd = EG_DMA_COPY_TILED;
 	lbpp = util_logbase2(bpp);
-	pitch_tile_max = ((pitch / bpp) >> 3) - 1;
+	pitch_tile_max = ((pitch / bpp) / 8) - 1;
 	nbanks = eg_num_banks(rctx->screen->b.tiling_info.num_banks);
 
 	if (dst_mode == RADEON_SURF_MODE_LINEAR) {
 		/* T2L */
 		array_mode = evergreen_array_mode(src_mode);
-		slice_tile_max = (rsrc->surface.level[src_level].nblk_x * rsrc->surface.level[src_level].nblk_y) >> 6;
+		slice_tile_max = (rsrc->surface.level[src_level].nblk_x * rsrc->surface.level[src_level].nblk_y) / (8*8);
 		slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
 		/* linear height must be the same as the slice tile max height, it's ok even
 		 * if the linear destination/source have smaller heigh as the size of the
@@ -3266,7 +3266,7 @@ static void evergreen_dma_copy_tile(struct r600_context *rctx,
 	} else {
 		/* L2T */
 		array_mode = evergreen_array_mode(dst_mode);
-		slice_tile_max = (rdst->surface.level[dst_level].nblk_x * rdst->surface.level[dst_level].nblk_y) >> 6;
+		slice_tile_max = (rdst->surface.level[dst_level].nblk_x * rdst->surface.level[dst_level].nblk_y) / (8*8);
 		slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
 		/* linear height must be the same as the slice tile max height, it's ok even
 		 * if the linear destination/source have smaller heigh as the size of the
@@ -3290,16 +3290,16 @@ static void evergreen_dma_copy_tile(struct r600_context *rctx,
 		addr += r600_resource_va(&rctx->screen->b.b, src);
 	}
 
-	size = (copy_height * pitch) >> 2;
-	ncopy = (size / 0x000fffff) + !!(size % 0x000fffff);
+	size = (copy_height * pitch) / 4;
+	ncopy = (size / EG_DMA_COPY_MAX_SIZE) + !!(size % EG_DMA_COPY_MAX_SIZE);
 	r600_need_dma_space(&rctx->b, ncopy * 9);
 
 	for (i = 0; i < ncopy; i++) {
 		cheight = copy_height;
-		if (((cheight * pitch) >> 2) > 0x000fffff) {
-			cheight = (0x000fffff << 2) / pitch;
+		if (((cheight * pitch) / 4) > EG_DMA_COPY_MAX_SIZE) {
+			cheight = (EG_DMA_COPY_MAX_SIZE * 4) / pitch;
 		}
-		size = (cheight * pitch) >> 2;
+		size = (cheight * pitch) / 4;
 		/* emit reloc before writting cs so that cs is always in consistent state */
 		r600_context_bo_reloc(&rctx->b, &rctx->b.rings.dma, &rsrc->resource,
 				      RADEON_USAGE_READ, RADEON_PRIO_MIN);
@@ -3381,7 +3381,7 @@ static void evergreen_dma_blit(struct pipe_context *ctx,
 	/* the x test here are currently useless (because we don't support partial blit)
 	 * but keep them around so we don't forget about those
 	 */
-	if ((src_pitch & 0x7) || (src_box->x & 0x7) || (dst_x & 0x7) || (src_box->y & 0x7) || (dst_y & 0x7)) {
+	if (src_pitch % 8 || src_box->x % 8 || dst_x % 8 || src_box->y % 8 || dst_y % 8) {
 		goto fallback;
 	}
 
diff --git a/src/gallium/drivers/r600/evergreend.h b/src/gallium/drivers/r600/evergreend.h
index 986fd16f1ba..9fde18432fc 100644
--- a/src/gallium/drivers/r600/evergreend.h
+++ b/src/gallium/drivers/r600/evergreend.h
@@ -2445,6 +2445,10 @@
 /* async DMA Packet types */
 #define    DMA_PACKET_WRITE                     0x2
 #define    DMA_PACKET_COPY                      0x3
+#define    EG_DMA_COPY_MAX_SIZE			0xfffff
+#define    EG_DMA_COPY_DWORD_ALIGNED		0x00
+#define    EG_DMA_COPY_BYTE_ALIGNED		0x40
+#define    EG_DMA_COPY_TILED			0x8
 #define    DMA_PACKET_INDIRECT_BUFFER           0x4
 #define    DMA_PACKET_SEMAPHORE                 0x5
 #define    DMA_PACKET_FENCE                     0x6
diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
index eb5500e29e5..267b326a511 100644
--- a/src/gallium/drivers/r600/r600_hw_context.c
+++ b/src/gallium/drivers/r600/r600_hw_context.c
@@ -448,7 +448,7 @@ void r600_dma_copy(struct r600_context *rctx,
 		uint64_t size)
 {
 	struct radeon_winsys_cs *cs = rctx->b.rings.dma.cs;
-	unsigned i, ncopy, csize, shift;
+	unsigned i, ncopy, csize;
 	struct r600_resource *rdst = (struct r600_resource*)dst;
 	struct r600_resource *rsrc = (struct r600_resource*)src;
 
@@ -458,13 +458,12 @@ void r600_dma_copy(struct r600_context *rctx,
 	util_range_add(&rdst->valid_buffer_range, dst_offset,
 		       dst_offset + size);
 
-	size >>= 2;
-	shift = 2;
-	ncopy = (size / 0xffff) + !!(size % 0xffff);
+	size >>= 2; /* convert to dwords */
+	ncopy = (size / R600_DMA_COPY_MAX_SIZE_DW) + !!(size % R600_DMA_COPY_MAX_SIZE_DW);
 
 	r600_need_dma_space(&rctx->b, ncopy * 5);
 	for (i = 0; i < ncopy; i++) {
-		csize = size < 0xffff ? size : 0xffff;
+		csize = size < R600_DMA_COPY_MAX_SIZE_DW ? size : R600_DMA_COPY_MAX_SIZE_DW;
 		/* emit reloc before writting cs so that cs is always in consistent state */
 		r600_context_bo_reloc(&rctx->b, &rctx->b.rings.dma, rsrc, RADEON_USAGE_READ,
 				      RADEON_PRIO_MIN);
@@ -475,8 +474,8 @@ void r600_dma_copy(struct r600_context *rctx,
 		cs->buf[cs->cdw++] = src_offset & 0xfffffffc;
 		cs->buf[cs->cdw++] = (dst_offset >> 32UL) & 0xff;
 		cs->buf[cs->cdw++] = (src_offset >> 32UL) & 0xff;
-		dst_offset += csize << shift;
-		src_offset += csize << shift;
+		dst_offset += csize << 2;
+		src_offset += csize << 2;
 		size -= csize;
 	}
 }
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index 4aa379826fd..d87caf89669 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -2802,12 +2802,12 @@ static boolean r600_dma_copy_tile(struct r600_context *rctx,
 
 	y = 0;
 	lbpp = util_logbase2(bpp);
-	pitch_tile_max = ((pitch / bpp) >> 3) - 1;
+	pitch_tile_max = ((pitch / bpp) / 8) - 1;
 
 	if (dst_mode == RADEON_SURF_MODE_LINEAR) {
 		/* T2L */
 		array_mode = r600_array_mode(src_mode);
-		slice_tile_max = (rsrc->surface.level[src_level].nblk_x * rsrc->surface.level[src_level].nblk_y) >> 6;
+		slice_tile_max = (rsrc->surface.level[src_level].nblk_x * rsrc->surface.level[src_level].nblk_y) / (8*8);
 		slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
 		/* linear height must be the same as the slice tile max height, it's ok even
 		 * if the linear destination/source have smaller heigh as the size of the
@@ -2826,7 +2826,7 @@ static boolean r600_dma_copy_tile(struct r600_context *rctx,
 	} else {
 		/* L2T */
 		array_mode = r600_array_mode(dst_mode);
-		slice_tile_max = (rdst->surface.level[dst_level].nblk_x * rdst->surface.level[dst_level].nblk_y) >> 6;
+		slice_tile_max = (rdst->surface.level[dst_level].nblk_x * rdst->surface.level[dst_level].nblk_y) / (8*8);
 		slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
 		/* linear height must be the same as the slice tile max height, it's ok even
 		 * if the linear destination/source have smaller heigh as the size of the
@@ -2844,20 +2844,20 @@ static boolean r600_dma_copy_tile(struct r600_context *rctx,
 		addr += src_y * pitch + src_x * bpp;
 	}
 	/* check that we are in dw/base alignment constraint */
-	if ((addr & 0x3) || (base & 0xff)) {
+	if (addr % 4 || base % 256) {
 		return FALSE;
 	}
 
 	/* It's a r6xx/r7xx limitation, the blit must be on 8 boundary for number
 	 * line in the blit. Compute max 8 line we can copy in the size limit
 	 */
-	cheight = ((0x0000ffff << 2) / pitch) & 0xfffffff8;
+	cheight = ((R600_DMA_COPY_MAX_SIZE_DW * 4) / pitch) & 0xfffffff8;
 	ncopy = (copy_height / cheight) + !!(copy_height % cheight);
 	r600_need_dma_space(&rctx->b, ncopy * 7);
 
 	for (i = 0; i < ncopy; i++) {
 		cheight = cheight > copy_height ? copy_height : cheight;
-		size = (cheight * pitch) >> 2;
+		size = (cheight * pitch) / 4;
 		/* emit reloc before writting cs so that cs is always in consistent state */
 		r600_context_bo_reloc(&rctx->b, &rctx->b.rings.dma, &rsrc->resource, RADEON_USAGE_READ,
 				      RADEON_PRIO_MIN);
@@ -2930,11 +2930,11 @@ static void r600_dma_blit(struct pipe_context *ctx,
 	dst_mode = dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED ? RADEON_SURF_MODE_LINEAR : dst_mode;
 
 	if (src_pitch != dst_pitch || src_box->x || dst_x || src_w != dst_w) {
-		/* strick requirement on r6xx/r7xx */
+		/* strict requirement on r6xx/r7xx */
 		goto fallback;
 	}
 	/* lot of constraint on alignment this should capture them all */
-	if ((src_pitch & 0x7) || (src_box->y & 0x7) || (dst_y & 0x7)) {
+	if (src_pitch % 8 || src_box->y % 8 || dst_y % 8) {
 		goto fallback;
 	}
 
@@ -2954,7 +2954,7 @@ static void r600_dma_blit(struct pipe_context *ctx,
 		dst_offset += dst_y * dst_pitch + dst_x * bpp;
 		size = src_box->height * src_pitch;
 		/* must be dw aligned */
-		if ((dst_offset & 0x3) || (src_offset & 0x3) || (size & 0x3)) {
+		if (dst_offset % 4 || src_offset % 4 || size % 4) {
 			goto fallback;
 		}
 		r600_dma_copy(rctx, dst, src, dst_offset, src_offset, size);
diff --git a/src/gallium/drivers/r600/r600d.h b/src/gallium/drivers/r600/r600d.h
index 05d1f0ab34a..f787803a53e 100644
--- a/src/gallium/drivers/r600/r600d.h
+++ b/src/gallium/drivers/r600/r600d.h
@@ -3743,6 +3743,7 @@
 /* async DMA Packet types */
 #define DMA_PACKET_WRITE		0x2
 #define DMA_PACKET_COPY			0x3
+#define R600_DMA_COPY_MAX_SIZE_DW		0xffff
 #define DMA_PACKET_INDIRECT_BUFFER	0x4
 #define DMA_PACKET_SEMAPHORE		0x5
 #define DMA_PACKET_FENCE		0x6