radeonsi: use TC L2 for updating descriptors on CIK

This allows not flushing TC L2 on CIK later.

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

2 files changed, 10 insertions, 5 deletions

diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c
index f0e353a1db9..e2da476ab0f 100644
--- a/src/gallium/drivers/radeonsi/si_descriptors.c
+++ b/src/gallium/drivers/radeonsi/si_descriptors.c
@@ -54,6 +54,7 @@ static uint32_t null_desc[8]; /* zeros */
  * packet. It's for preventing a read-after-write (RAW) hazard between two
  * CP DMA packets. */
 #define SI_CP_DMA_RAW_WAIT	(1 << 1) /* SI+ */
+#define CIK_CP_DMA_USE_L2	(1 << 2)
 
 /* Emit a CP DMA packet to do a copy from one buffer to another.
  * The size must fit in bits [20:0].
@@ -65,13 +66,15 @@ static void si_emit_cp_dma_copy_buffer(struct si_context *sctx,
 	struct radeon_winsys_cs *cs = sctx->b.rings.gfx.cs;
 	uint32_t sync_flag = flags & R600_CP_DMA_SYNC ? PKT3_CP_DMA_CP_SYNC : 0;
 	uint32_t raw_wait = flags & SI_CP_DMA_RAW_WAIT ? PKT3_CP_DMA_CMD_RAW_WAIT : 0;
+	uint32_t sel = flags & CIK_CP_DMA_USE_L2 ?
+			   PKT3_CP_DMA_SRC_SEL(3) | PKT3_CP_DMA_DST_SEL(3) : 0;
 
 	assert(size);
 	assert((size & ((1<<21)-1)) == size);
 
 	if (sctx->b.chip_class >= CIK) {
 		radeon_emit(cs, PKT3(PKT3_DMA_DATA, 5, 0));
-		radeon_emit(cs, sync_flag);		/* CP_SYNC [31] */
+		radeon_emit(cs, sync_flag | sel);	/* CP_SYNC [31] */
 		radeon_emit(cs, src_va);		/* SRC_ADDR_LO [31:0] */
 		radeon_emit(cs, src_va >> 32);		/* SRC_ADDR_HI [31:0] */
 		radeon_emit(cs, dst_va);		/* DST_ADDR_LO [31:0] */
@@ -95,13 +98,14 @@ static void si_emit_cp_dma_clear_buffer(struct si_context *sctx,
 	struct radeon_winsys_cs *cs = sctx->b.rings.gfx.cs;
 	uint32_t sync_flag = flags & R600_CP_DMA_SYNC ? PKT3_CP_DMA_CP_SYNC : 0;
 	uint32_t raw_wait = flags & SI_CP_DMA_RAW_WAIT ? PKT3_CP_DMA_CMD_RAW_WAIT : 0;
+	uint32_t dst_sel = flags & CIK_CP_DMA_USE_L2 ? PKT3_CP_DMA_DST_SEL(3) : 0;
 
 	assert(size);
 	assert((size & ((1<<21)-1)) == size);
 
 	if (sctx->b.chip_class >= CIK) {
 		radeon_emit(cs, PKT3(PKT3_DMA_DATA, 5, 0));
-		radeon_emit(cs, sync_flag | PKT3_CP_DMA_SRC_SEL(2)); /* CP_SYNC [31] | SRC_SEL[30:29] */
+		radeon_emit(cs, sync_flag | dst_sel | PKT3_CP_DMA_SRC_SEL(2)); /* CP_SYNC [31] | SRC_SEL[30:29] */
 		radeon_emit(cs, clear_value);		/* DATA [31:0] */
 		radeon_emit(cs, 0);
 		radeon_emit(cs, dst_va);		/* DST_ADDR_LO [31:0] */
@@ -145,7 +149,7 @@ static void si_init_descriptors(struct si_context *sctx,
 	 * only once at context initialization. */
 	si_emit_cp_dma_clear_buffer(sctx, desc->buffer->gpu_address,
 				    desc->buffer->b.b.width0, 0,
-				    R600_CP_DMA_SYNC);
+				    R600_CP_DMA_SYNC | CIK_CP_DMA_USE_L2);
 }
 
 static void si_release_descriptors(struct si_descriptors *desc)
@@ -227,11 +231,10 @@ static void si_emit_descriptors(struct si_context *sctx,
 	va_base = desc->buffer->gpu_address;
 
 	/* Copy the descriptors to a new context slot. */
-	/* XXX Consider using TC or L2 for this copy on CIK. */
 	si_emit_cp_dma_copy_buffer(sctx,
 				   va_base + new_context_id * desc->context_size,
 				   va_base + desc->current_context_id * desc->context_size,
-				   desc->context_size, R600_CP_DMA_SYNC);
+				   desc->context_size, R600_CP_DMA_SYNC | CIK_CP_DMA_USE_L2);
 
 	va_base += new_context_id * desc->context_size;
 
diff --git a/src/gallium/drivers/radeonsi/sid.h b/src/gallium/drivers/radeonsi/sid.h
index e36244dc7d4..afe011b15c7 100644
--- a/src/gallium/drivers/radeonsi/sid.h
+++ b/src/gallium/drivers/radeonsi/sid.h
@@ -164,10 +164,12 @@
 /* 0 - SRC_ADDR
  * 1 - GDS (program SAS to 1 as well)
  * 2 - DATA
+ * 3 - SRC_ADDR using TC L2 (DMA_DATA only)
  */
 #define PKT3_CP_DMA_DST_SEL(x)       ((x) << 20)
 /* 0 - DST_ADDR
  * 1 - GDS (program DAS to 1 as well)
+ * 3 - DST_ADDR using TC L2 (DMA_DATA only)
  */
 /* COMMAND */
 #define PKT3_CP_DMA_CMD_SRC_SWAP(x) ((x) << 23)