freedreno/ir3: add ctx->mem_to_mem()

For dealing with indirect-draw + gl_VertexID, we'll introduce another
case where we need to use CP_MEM_TO_MEM.  Rather than adding more
if(a5xx)/else make this a ctx vfunc.

Signed-off-by: Rob Clark <[email protected]>

4 files changed, 49 insertions, 14 deletions

diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
index dca3692b047..5fec2b6b08a 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
@@ -912,6 +912,26 @@ fd4_emit_ib(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
 	__OUT_IB(ring, true, target);
 }
 
+static void
+fd4_mem_to_mem(struct fd_ringbuffer *ring, struct pipe_resource *dst,
+		unsigned dst_off, struct pipe_resource *src, unsigned src_off,
+		unsigned sizedwords)
+{
+	struct fd_bo *src_bo = fd_resource(src)->bo;
+	struct fd_bo *dst_bo = fd_resource(dst)->bo;
+	unsigned i;
+
+	for (i = 0; i < sizedwords; i++) {
+		OUT_PKT3(ring, CP_MEM_TO_MEM, 3);
+		OUT_RING(ring, 0x00000000);
+		OUT_RELOCW(ring, dst_bo, dst_off, 0, 0);
+		OUT_RELOC (ring, src_bo, src_off, 0, 0);
+
+		dst_off += 4;
+		src_off += 4;
+	}
+}
+
 void
 fd4_emit_init(struct pipe_context *pctx)
 {
@@ -919,4 +939,5 @@ fd4_emit_init(struct pipe_context *pctx)
 	ctx->emit_const = fd4_emit_const;
 	ctx->emit_const_bo = fd4_emit_const_bo;
 	ctx->emit_ib = fd4_emit_ib;
+	ctx->mem_to_mem = fd4_mem_to_mem;
 }
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_emit.c b/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
index 08e9bb7bafb..b7ce084a827 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
@@ -1068,6 +1068,26 @@ fd5_emit_ib(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
 	__OUT_IB5(ring, target);
 }
 
+static void
+fd5_mem_to_mem(struct fd_ringbuffer *ring, struct pipe_resource *dst,
+		unsigned dst_off, struct pipe_resource *src, unsigned src_off,
+		unsigned sizedwords)
+{
+	struct fd_bo *src_bo = fd_resource(src)->bo;
+	struct fd_bo *dst_bo = fd_resource(dst)->bo;
+	unsigned i;
+
+	for (i = 0; i < sizedwords; i++) {
+		OUT_PKT7(ring, CP_MEM_TO_MEM, 5);
+		OUT_RING(ring, 0x00000000);
+		OUT_RELOCW(ring, dst_bo, dst_off, 0, 0);
+		OUT_RELOC (ring, src_bo, src_off, 0, 0);
+
+		dst_off += 4;
+		src_off += 4;
+	}
+}
+
 void
 fd5_emit_init(struct pipe_context *pctx)
 {
@@ -1075,4 +1095,5 @@ fd5_emit_init(struct pipe_context *pctx)
 	ctx->emit_const = fd5_emit_const;
 	ctx->emit_const_bo = fd5_emit_const_bo;
 	ctx->emit_ib = fd5_emit_ib;
+	ctx->mem_to_mem = fd5_mem_to_mem;
 }
diff --git a/src/gallium/drivers/freedreno/freedreno_context.h b/src/gallium/drivers/freedreno/freedreno_context.h
index 1e9911ea9bf..a4e1e4bb86c 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.h
+++ b/src/gallium/drivers/freedreno/freedreno_context.h
@@ -332,6 +332,11 @@ struct fd_context {
 	/* blit: */
 	void (*blit)(struct fd_context *ctx, const struct pipe_blit_info *info);
 
+	/* simple gpu "memcpy": */
+	void (*mem_to_mem)(struct fd_ringbuffer *ring, struct pipe_resource *dst,
+			unsigned dst_off, struct pipe_resource *src, unsigned src_off,
+			unsigned sizedwords);
+
 	/*
 	 * Common pre-cooked VBO state (used for a3xx and later):
 	 */
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.c b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
index 3337543113d..3b1fcdfd5f9 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
@@ -879,20 +879,8 @@ ir3_emit_cs_consts(const struct ir3_shader_variant *v, struct fd_ringbuffer *rin
 					0x1000);
 				indirect_offset = 0;
 
-				if (is_a5xx(ctx->screen)) {
-					struct fd_bo *src = fd_resource(info->indirect)->bo;
-					struct fd_bo *dst = fd_resource(indirect)->bo;
-					for (unsigned i = 0; i < 3; i++) {
-						unsigned dst_off = i * 4;
-						unsigned src_off = (i * 4) + info->indirect_offset;
-						OUT_PKT7(ring, CP_MEM_TO_MEM, 5);
-						OUT_RING(ring, 0x00000000);
-						OUT_RELOCW(ring, dst, dst_off, 0, 0);
-						OUT_RELOC (ring, src, src_off, 0, 0);
-					}
-				} else {
-					assert(0);
-				}
+				ctx->mem_to_mem(ring, indirect, 0, info->indirect,
+						info->indirect_offset, 3);
 			} else {
 				pipe_resource_reference(&indirect, info->indirect);
 				indirect_offset = info->indirect_offset;