radeonsi: fix and enable transform feedback for CIK

The CP_STRMOUT_CNTL register was moved again.

Reviewed-by: Alex Deucher <[email protected]>

4 files changed, 40 insertions, 28 deletions

diff --git a/src/gallium/drivers/radeon/r600_cs.h b/src/gallium/drivers/radeon/r600_cs.h
index f56c92c1017..fa749dad6c9 100644
--- a/src/gallium/drivers/radeon/r600_cs.h
+++ b/src/gallium/drivers/radeon/r600_cs.h
@@ -94,4 +94,18 @@ static INLINE void r600_write_context_reg(struct radeon_winsys_cs *cs, unsigned
 	radeon_emit(cs, value);
 }
 
+static INLINE void cik_write_uconfig_reg_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
+{
+	assert(reg >= CIK_UCONFIG_REG_OFFSET && reg < CIK_UCONFIG_REG_END);
+	assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
+	radeon_emit(cs, PKT3(PKT3_SET_UCONFIG_REG, num, 0));
+	radeon_emit(cs, (reg - CIK_UCONFIG_REG_OFFSET) >> 2);
+}
+
+static INLINE void cik_write_uconfig_reg(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
+{
+	cik_write_uconfig_reg_seq(cs, reg, 1);
+	radeon_emit(cs, value);
+}
+
 #endif
diff --git a/src/gallium/drivers/radeon/r600_streamout.c b/src/gallium/drivers/radeon/r600_streamout.c
index 18f7d8883b5..0f65de4a805 100644
--- a/src/gallium/drivers/radeon/r600_streamout.c
+++ b/src/gallium/drivers/radeon/r600_streamout.c
@@ -145,39 +145,35 @@ void r600_set_streamout_targets(struct pipe_context *ctx,
 static void r600_flush_vgt_streamout(struct r600_common_context *rctx)
 {
 	struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
+	unsigned reg_strmout_cntl;
 
-	r600_write_config_reg(cs, R_008490_CP_STRMOUT_CNTL, 0);
+	/* The register is at different places on different ASICs. */
+	if (rctx->chip_class >= CIK) {
+		reg_strmout_cntl = R_0300FC_CP_STRMOUT_CNTL;
+	} else if (rctx->chip_class >= EVERGREEN) {
+		reg_strmout_cntl = R_0084FC_CP_STRMOUT_CNTL;
+	} else {
+		reg_strmout_cntl = R_008490_CP_STRMOUT_CNTL;
+	}
+
+	if (rctx->chip_class >= CIK) {
+		cik_write_uconfig_reg(cs, reg_strmout_cntl, 0);
+	} else {
+		r600_write_config_reg(cs, reg_strmout_cntl, 0);
+	}
 
 	radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
 	radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0));
 
 	radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, 0));
 	radeon_emit(cs, WAIT_REG_MEM_EQUAL); /* wait until the register is equal to the reference value */
-	radeon_emit(cs, R_008490_CP_STRMOUT_CNTL >> 2);  /* register */
+	radeon_emit(cs, reg_strmout_cntl >> 2);  /* register */
 	radeon_emit(cs, 0);
 	radeon_emit(cs, S_008490_OFFSET_UPDATE_DONE(1)); /* reference value */
 	radeon_emit(cs, S_008490_OFFSET_UPDATE_DONE(1)); /* mask */
 	radeon_emit(cs, 4); /* poll interval */
 }
 
-static void evergreen_flush_vgt_streamout(struct r600_common_context *rctx)
-{
-	struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
-
-	r600_write_config_reg(cs, R_0084FC_CP_STRMOUT_CNTL, 0);
-
-	radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
-	radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0));
-
-	radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, 0));
-	radeon_emit(cs, WAIT_REG_MEM_EQUAL); /* wait until the register is equal to the reference value */
-	radeon_emit(cs, R_0084FC_CP_STRMOUT_CNTL >> 2);  /* register */
-	radeon_emit(cs, 0);
-	radeon_emit(cs, S_0084FC_OFFSET_UPDATE_DONE(1)); /* reference value */
-	radeon_emit(cs, S_0084FC_OFFSET_UPDATE_DONE(1)); /* mask */
-	radeon_emit(cs, 4); /* poll interval */
-}
-
 static void r600_set_streamout_enable(struct r600_common_context *rctx, unsigned buffer_enable_bit)
 {
 	struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
@@ -224,11 +220,11 @@ static void r600_emit_streamout_begin(struct r600_common_context *rctx, struct r
 	unsigned *stride_in_dw = rctx->streamout.stride_in_dw;
 	unsigned i, update_flags = 0;
 
+	r600_flush_vgt_streamout(rctx);
+
 	if (rctx->chip_class >= EVERGREEN) {
-		evergreen_flush_vgt_streamout(rctx);
 		evergreen_set_streamout_enable(rctx, rctx->streamout.enabled_mask);
 	} else {
-		r600_flush_vgt_streamout(rctx);
 		r600_set_streamout_enable(rctx, rctx->streamout.enabled_mask);
 	}
 
@@ -315,11 +311,7 @@ void r600_emit_streamout_end(struct r600_common_context *rctx)
 	unsigned i;
 	uint64_t va;
 
-	if (rctx->chip_class >= EVERGREEN) {
-		evergreen_flush_vgt_streamout(rctx);
-	} else {
-		r600_flush_vgt_streamout(rctx);
-	}
+	r600_flush_vgt_streamout(rctx);
 
 	for (i = 0; i < rctx->streamout.num_targets; i++) {
 		if (!t[i])
diff --git a/src/gallium/drivers/radeon/r600d_common.h b/src/gallium/drivers/radeon/r600d_common.h
index b179757b424..99146941570 100644
--- a/src/gallium/drivers/radeon/r600d_common.h
+++ b/src/gallium/drivers/radeon/r600d_common.h
@@ -28,6 +28,8 @@
 
 #define R600_CONFIG_REG_OFFSET	0x08000
 #define R600_CONTEXT_REG_OFFSET 0x28000
+#define CIK_UCONFIG_REG_OFFSET               0x00030000
+#define CIK_UCONFIG_REG_END                  0x00031000
 
 #define PKT_TYPE_S(x)                   (((x) & 0x3) << 30)
 #define PKT_COUNT_S(x)                  (((x) & 0x3FFF) << 16)
@@ -57,6 +59,7 @@
 #define		SURFACE_BASE_UPDATE_COLOR(x)   (2 << (x))
 #define		SURFACE_BASE_UPDATE_COLOR_NUM(x) (((1 << x) - 1) << 1)
 #define		SURFACE_BASE_UPDATE_STRMOUT(x) (0x200 << (x))
+#define PKT3_SET_UCONFIG_REG                   0x79 /* new for CIK */
 
 #define EVENT_TYPE_PS_PARTIAL_FLUSH            0x10
 #define EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT 0x14
@@ -140,4 +143,7 @@
 #define   G_028B98_STREAM_3_BUFFER_EN(x)                              (((x) >> 12) & 0x0F)
 #define   C_028B98_STREAM_3_BUFFER_EN                                 0xFFFF0FFF
 
+/*CIK+*/
+#define R_0300FC_CP_STRMOUT_CNTL		     0x0300FC
+
 #endif
diff --git a/src/gallium/drivers/radeonsi/radeonsi_pipe.c b/src/gallium/drivers/radeonsi/radeonsi_pipe.c
index 5220e4184b8..138268ca418 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_pipe.c
+++ b/src/gallium/drivers/radeonsi/radeonsi_pipe.c
@@ -342,7 +342,7 @@ static const char* r600_get_name(struct pipe_screen* pscreen)
 static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 {
 	struct r600_screen *rscreen = (struct r600_screen *)pscreen;
-	bool has_streamout = HAVE_LLVM >= 0x0304 && rscreen->b.chip_class == SI;
+	bool has_streamout = HAVE_LLVM >= 0x0304;
 
 	switch (param) {
 	/* Supported features (boolean caps). */