amd: GFX9 packet changes

Reviewed-by: Nicolai Hähnle <[email protected]>

4 files changed, 26 insertions, 13 deletions

diff --git a/src/amd/common/r600d_common.h b/src/amd/common/r600d_common.h
index a35108fc0cd..3fdfb7c695f 100644
--- a/src/amd/common/r600d_common.h
+++ b/src/amd/common/r600d_common.h
@@ -63,6 +63,7 @@
 		 * 3 - send 64bit GPU counter value
 		 * 4 - send 64bit sys counter value
 		 */
+#define PKT3_RELEASE_MEM                       0x49 /* GFX9+ */
 #define PKT3_SET_CONFIG_REG		       0x68
 #define PKT3_SET_CONTEXT_REG		       0x69
 #define PKT3_STRMOUT_BASE_UPDATE	       0x72 /* r700 only */
diff --git a/src/amd/common/sid.h b/src/amd/common/sid.h
index 61e14067ce9..e0c3a02ddcd 100644
--- a/src/amd/common/sid.h
+++ b/src/amd/common/sid.h
@@ -50,7 +50,16 @@
 		 * 4 - *S_PARTIAL_FLUSH
 		 * 5 - TS events
 		 */
-#define EVENT_WRITE_INV_L2                   0x100000
+
+/* EVENT_WRITE_EOP (SI-VI) & RELEASE_MEM (GFX9) */
+#define EVENT_TCL1_VOL_ACTION_ENA		(1 << 12)
+#define EVENT_TC_VOL_ACTION_ENA			(1 << 13)
+#define EVENT_TC_WB_ACTION_ENA			(1 << 15)
+#define EVENT_TCL1_ACTION_ENA			(1 << 16)
+#define EVENT_TC_ACTION_ENA			(1 << 17)
+#define EVENT_TC_NC_ACTION_ENA			(1 << 19) /* GFX9+ */
+#define EVENT_TC_WC_ACTION_ENA			(1 << 20) /* GFX9+ */
+#define EVENT_TC_MD_ACTION_ENA			(1 << 21) /* GFX9+ */
 
 
 #define PREDICATION_OP_CLEAR 0x0
@@ -92,7 +101,7 @@
 #define     CONTEXT_CONTROL_LOAD_ENABLE(x)     (((unsigned)(x) & 0x1) << 31)
 #define     CONTEXT_CONTROL_LOAD_CE_RAM(x)     (((unsigned)(x) & 0x1) << 28)
 #define     CONTEXT_CONTROL_SHADOW_ENABLE(x)   (((unsigned)(x) & 0x1) << 31)
-#define PKT3_INDEX_TYPE                        0x2A
+#define PKT3_INDEX_TYPE                        0x2A /* not on GFX9 */
 #define PKT3_DRAW_INDIRECT_MULTI               0x2C
 #define   R_2C3_DRAW_INDEX_LOC                  0x2C3
 #define     S_2C3_COUNT_INDIRECT_ENABLE(x)      (((unsigned)(x) & 0x1) << 30)
@@ -153,13 +162,14 @@
 #define PKT3_ME_INITIALIZE                     0x44 /* not on CIK */
 #define PKT3_COND_WRITE                        0x45
 #define PKT3_EVENT_WRITE                       0x46
-#define PKT3_EVENT_WRITE_EOP                   0x47
+#define PKT3_EVENT_WRITE_EOP                   0x47 /* not on GFX9 */
 /* CP DMA bug: Any use of CP_DMA.DST_SEL=TC must be avoided when EOS packets
  * are used. Use DST_SEL=MC instead. For prefetch, use SRC_SEL=TC and
  * DST_SEL=MC. Only CIK chips are affected.
  */
-/*#define PKT3_EVENT_WRITE_EOS                   0x48*/ /* fix CP DMA before uncommenting */
-#define PKT3_RELEASE_MEM                       0x49
+/* fix CP DMA before uncommenting: */
+/*#define PKT3_EVENT_WRITE_EOS                   0x48*/ /* not on GFX9 */
+#define PKT3_RELEASE_MEM                       0x49 /* GFX9+ (any ring) or GFX8 (compute ring only) */
 #define PKT3_ONE_REG_WRITE                     0x57 /* not on CIK */
 #define PKT3_ACQUIRE_MEM                       0x58 /* new for CIK */
 #define PKT3_SET_CONFIG_REG                    0x68
@@ -221,14 +231,15 @@
 #define   R_413_CP_DMA_WORD3		0x413 /* 0x[packet number][word index] */
 #define     S_413_DST_ADDR_HI(x)	((x) & 0xffff)
 #define   R_414_COMMAND			0x414
-#define     S_414_BYTE_COUNT(x)		((x) & 0x1fffff)
-#define     S_414_DISABLE_WR_CONFIRM(x)	(((unsigned)(x) & 0x1) << 21)
-#define     S_414_SRC_SWAP(x)		(((unsigned)(x) & 0x3) << 22)
+#define     S_414_BYTE_COUNT_GFX6(x)	((x) & 0x1fffff)
+#define     S_414_BYTE_COUNT_GFX9(x)	((x) & 0x3ffffff)
+#define     S_414_DISABLE_WR_CONFIRM_GFX6(x) (((unsigned)(x) & 0x1) << 21) /* not on GFX9 */
+#define     S_414_SRC_SWAP(x)		(((unsigned)(x) & 0x3) << 22) /* not on GFX9 */
 #define       V_414_NONE		0
 #define       V_414_8_IN_16		1
 #define       V_414_8_IN_32		2
 #define       V_414_8_IN_64		3
-#define     S_414_DST_SWAP(x)		(((unsigned)(x) & 0x3) << 24)
+#define     S_414_DST_SWAP(x)		(((unsigned)(x) & 0x3) << 24) /* not on GFX9 */
 #define       V_414_NONE		0
 #define       V_414_8_IN_16		1
 #define       V_414_8_IN_32		2
@@ -246,6 +257,7 @@
 #define       V_414_INCREMENT		0
 #define       V_414_NO_INCREMENT	1
 #define     S_414_RAW_WAIT(x)		(((unsigned)(x) & 0x1) << 30)
+#define     S_414_DISABLE_WR_CONFIRM_GFX9(x) (((unsigned)(x) & 0x1) << 31)
 
 #define PKT3_DMA_DATA					0x50 /* new for CIK */
 /* 1. header
diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index 6e50f64a29a..49c99bcab77 100644
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -869,7 +869,7 @@ static void si_emit_cp_dma_copy_buffer(struct radv_cmd_buffer *cmd_buffer,
 {
 	struct radeon_winsys_cs *cs = cmd_buffer->cs;
 	uint32_t sync_flag = flags & R600_CP_DMA_SYNC ? S_411_CP_SYNC(1) : 0;
-	uint32_t wr_confirm = !(flags & R600_CP_DMA_SYNC) ? S_414_DISABLE_WR_CONFIRM(1) : 0;
+	uint32_t wr_confirm = !(flags & R600_CP_DMA_SYNC) ? S_414_DISABLE_WR_CONFIRM_GFX6(1) : 0;
 	uint32_t raw_wait = flags & SI_CP_DMA_RAW_WAIT ? S_414_RAW_WAIT(1) : 0;
 	uint32_t sel = flags & CIK_CP_DMA_USE_L2 ?
 			   S_411_SRC_SEL(V_411_SRC_ADDR_TC_L2) |
@@ -917,7 +917,7 @@ static void si_emit_cp_dma_clear_buffer(struct radv_cmd_buffer *cmd_buffer,
 {
 	struct radeon_winsys_cs *cs = cmd_buffer->cs;
 	uint32_t sync_flag = flags & R600_CP_DMA_SYNC ? S_411_CP_SYNC(1) : 0;
-	uint32_t wr_confirm = !(flags & R600_CP_DMA_SYNC) ? S_414_DISABLE_WR_CONFIRM(1) : 0;
+	uint32_t wr_confirm = !(flags & R600_CP_DMA_SYNC) ? S_414_DISABLE_WR_CONFIRM_GFX6(1) : 0;
 	uint32_t raw_wait = flags & SI_CP_DMA_RAW_WAIT ? S_414_RAW_WAIT(1) : 0;
 	uint32_t dst_sel = flags & CIK_CP_DMA_USE_L2 ? S_411_DSL_SEL(V_411_DST_ADDR_TC_L2) : 0;
 
diff --git a/src/gallium/drivers/radeonsi/si_cp_dma.c b/src/gallium/drivers/radeonsi/si_cp_dma.c
index 812fcbc2b21..9d236daf731 100644
--- a/src/gallium/drivers/radeonsi/si_cp_dma.c
+++ b/src/gallium/drivers/radeonsi/si_cp_dma.c
@@ -51,7 +51,7 @@ static void si_emit_cp_dma(struct si_context *sctx, uint64_t dst_va,
 			   enum r600_coherency coher)
 {
 	struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
-	uint32_t header = 0, command = S_414_BYTE_COUNT(size);
+	uint32_t header = 0, command = S_414_BYTE_COUNT_GFX6(size);
 
 	assert(size);
 	assert(size <= CP_DMA_MAX_BYTE_COUNT);
@@ -60,7 +60,7 @@ static void si_emit_cp_dma(struct si_context *sctx, uint64_t dst_va,
 	if (flags & CP_DMA_SYNC)
 		header |= S_411_CP_SYNC(1);
 	else
-		command |= S_414_DISABLE_WR_CONFIRM(1);
+		command |= S_414_DISABLE_WR_CONFIRM_GFX6(1);
 
 	if (flags & CP_DMA_RAW_WAIT)
 		command |= S_414_RAW_WAIT(1);