freedreno/a6xx: cleanup magic registers

Extract out values for the handful of unknown registers which have
different values across different a6xx models, to simplify adding
support for new a6xx's.

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

6 files changed, 47 insertions, 22 deletions

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_blitter.c b/src/gallium/drivers/freedreno/a6xx/fd6_blitter.c
index 9e145e55e5b..67924a8185f 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_blitter.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_blitter.c
@@ -195,7 +195,8 @@ blit_control(enum a6xx_color_fmt fmt)
  * supported by hw.. if necessary decompose into (potentially) two 2D blits
  */
 static void
-emit_blit_buffer(struct fd_ringbuffer *ring, const struct pipe_blit_info *info)
+emit_blit_buffer(struct fd_context *ctx, struct fd_ringbuffer *ring,
+		const struct pipe_blit_info *info)
 {
 	const struct pipe_box *sbox = &info->src.box;
 	const struct pipe_box *dbox = &info->dst.box;
@@ -327,7 +328,7 @@ emit_blit_buffer(struct fd_ringbuffer *ring, const struct pipe_blit_info *info)
 		OUT_RING(ring, 0xf180);
 
 		OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
-		OUT_RING(ring, 0x01000000);
+		OUT_RING(ring, fd6_context(ctx)->magic.RB_UNKNOWN_8E04_blit);
 
 		OUT_PKT7(ring, CP_BLIT, 1);
 		OUT_RING(ring, CP_BLIT_0_OP(BLIT_OP_SCALE));
@@ -335,12 +336,13 @@ emit_blit_buffer(struct fd_ringbuffer *ring, const struct pipe_blit_info *info)
 		OUT_WFI5(ring);
 
 		OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
-		OUT_RING(ring, 0);
+		OUT_RING(ring, 0);             /* RB_UNKNOWN_8E04 */
 	}
 }
 
 static void
-emit_blit_texture(struct fd_ringbuffer *ring, const struct pipe_blit_info *info)
+emit_blit_texture(struct fd_context *ctx, struct fd_ringbuffer *ring,
+		const struct pipe_blit_info *info)
 {
 	const struct pipe_box *sbox = &info->src.box;
 	const struct pipe_box *dbox = &info->dst.box;
@@ -540,7 +542,7 @@ emit_blit_texture(struct fd_ringbuffer *ring, const struct pipe_blit_info *info)
 				0xf000);
 
 		OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
-		OUT_RING(ring, 0x01000000);
+		OUT_RING(ring, fd6_context(ctx)->magic.RB_UNKNOWN_8E04_blit);
 
 		OUT_PKT7(ring, CP_BLIT, 1);
 		OUT_RING(ring, CP_BLIT_0_OP(BLIT_OP_SCALE));
@@ -548,7 +550,7 @@ emit_blit_texture(struct fd_ringbuffer *ring, const struct pipe_blit_info *info)
 		OUT_WFI5(ring);
 
 		OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
-		OUT_RING(ring, 0);
+		OUT_RING(ring, 0);             /* RB_UNKNOWN_8E04 */
 	}
 }
 
@@ -676,12 +678,12 @@ handle_rgba_blit(struct fd_context *ctx, const struct pipe_blit_info *info)
 			(info->dst.resource->target == PIPE_BUFFER)) {
 		assert(fd_resource(info->src.resource)->tile_mode == TILE6_LINEAR);
 		assert(fd_resource(info->dst.resource)->tile_mode == TILE6_LINEAR);
-		emit_blit_buffer(batch->draw, info);
+		emit_blit_buffer(ctx, batch->draw, info);
 	} else {
 		/* I don't *think* we need to handle blits between buffer <-> !buffer */
 		debug_assert(info->src.resource->target != PIPE_BUFFER);
 		debug_assert(info->dst.resource->target != PIPE_BUFFER);
-		emit_blit_texture(batch->draw, info);
+		emit_blit_texture(ctx, batch->draw, info);
 	}
 
 	fd6_event_write(batch, batch->draw, 0x1d, true);
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_context.c b/src/gallium/drivers/freedreno/a6xx/fd6_context.c
index aa86af5c050..ec8b871a28f 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_context.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_context.c
@@ -90,6 +90,18 @@ fd6_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
 	if (!fd6_ctx)
 		return NULL;
 
+
+	switch (screen->gpu_id) {
+	case 630:
+		fd6_ctx->magic.RB_UNKNOWN_8E04_blit = 0x01000000;
+		// NOTE: newer blob using 0x3c400004, need to revisit:
+		fd6_ctx->magic.RB_CCU_CNTL_gmem     = 0x7c400004;
+		fd6_ctx->magic.RB_CCU_CNTL_bypass   = 0x10000000;
+		fd6_ctx->magic.PC_UNKNOWN_9805 = 0x1;
+		fd6_ctx->magic.SP_UNKNOWN_A0F8 = 0x1;
+		break;
+	}
+
 	pctx = &fd6_ctx->base.base;
 	pctx->screen = pscreen;
 
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_context.h b/src/gallium/drivers/freedreno/a6xx/fd6_context.h
index 76183cca14a..0d810d350e9 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_context.h
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_context.h
@@ -91,6 +91,18 @@ struct fd6_context {
 
 	uint16_t tex_seqno;
 	struct hash_table *tex_cache;
+
+	/* collection of magic register values which differ between
+	 * various different a6xx
+	 */
+	struct {
+		uint32_t RB_UNKNOWN_8E04_blit;    /* value for CP_BLIT's */
+		uint32_t RB_CCU_CNTL_bypass;      /* for sysmem rendering */
+		uint32_t RB_CCU_CNTL_gmem;        /* for GMEM rendering */
+		uint32_t PC_UNKNOWN_9805;
+		uint32_t SP_UNKNOWN_A0F8;
+
+	} magic;
 };
 
 static inline struct fd6_context *
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_draw.c b/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
index a0df3dbc1ea..fcf0dc19233 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
@@ -256,6 +256,7 @@ static void
 fd6_clear_lrz(struct fd_batch *batch, struct fd_resource *zsbuf, double depth)
 {
 	struct fd_ringbuffer *ring;
+	struct fd6_context *fd6_ctx = fd6_context(batch->ctx);
 
 	// TODO mid-frame clears (ie. app doing crazy stuff)??  Maybe worth
 	// splitting both clear and lrz clear out into their own rb's.  And
@@ -277,7 +278,7 @@ fd6_clear_lrz(struct fd_batch *batch, struct fd_resource *zsbuf, double depth)
 	OUT_WFI5(ring);
 
 	OUT_PKT4(ring, REG_A6XX_RB_CCU_CNTL, 1);
-	OUT_RING(ring, 0x10000000);
+	OUT_RING(ring, fd6_ctx->magic.RB_CCU_CNTL_bypass);
 
 	OUT_PKT4(ring, REG_A6XX_HLSQ_UPDATE_CNTL, 1);
 	OUT_RING(ring, 0x7ffff);
@@ -354,7 +355,7 @@ fd6_clear_lrz(struct fd_batch *batch, struct fd_resource *zsbuf, double depth)
 	OUT_WFI5(ring);
 
 	OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
-	OUT_RING(ring, 0x1000000);
+	OUT_RING(ring, fd6_ctx->magic.RB_UNKNOWN_8E04_blit);
 
 	OUT_PKT7(ring, CP_BLIT, 1);
 	OUT_RING(ring, CP_BLIT_0_OP(BLIT_OP_SCALE));
@@ -362,7 +363,7 @@ fd6_clear_lrz(struct fd_batch *batch, struct fd_resource *zsbuf, double depth)
 	OUT_WFI5(ring);
 
 	OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
-	OUT_RING(ring, 0x0);
+	OUT_RING(ring, 0x0);               /* RB_UNKNOWN_8E04 */
 
 	fd6_event_write(batch, ring, UNK_1D, true);
 	fd6_event_write(batch, ring, FACENESS_FLUSH, true);
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
index 12ce00d976d..1cd85457a93 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
@@ -1218,8 +1218,7 @@ fd6_emit_restore(struct fd_batch *batch, struct fd_ringbuffer *ring)
 
 	OUT_WFI5(ring);
 
-	WRITE(REG_A6XX_RB_CCU_CNTL, 0x7c400004);
-	WRITE(REG_A6XX_RB_UNKNOWN_8E04, 0x00100000);
+	WRITE(REG_A6XX_RB_UNKNOWN_8E04, 0x0);
 	WRITE(REG_A6XX_SP_UNKNOWN_AE04, 0x8);
 	WRITE(REG_A6XX_SP_UNKNOWN_AE00, 0);
 	WRITE(REG_A6XX_SP_UNKNOWN_AE0F, 0x3f);
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_gmem.c b/src/gallium/drivers/freedreno/a6xx/fd6_gmem.c
index be8ea063812..6903d9c854e 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_gmem.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_gmem.c
@@ -642,6 +642,7 @@ emit_binning_pass(struct fd_batch *batch)
 {
 	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+	struct fd6_context *fd6_ctx = fd6_context(batch->ctx);
 
 	uint32_t x1 = gmem->minx;
 	uint32_t y1 = gmem->miny;
@@ -669,10 +670,10 @@ emit_binning_pass(struct fd_batch *batch)
 	update_vsc_pipe(batch);
 
 	OUT_PKT4(ring, REG_A6XX_PC_UNKNOWN_9805, 1);
-	OUT_RING(ring, 0x1);
+	OUT_RING(ring, fd6_ctx->magic.PC_UNKNOWN_9805);
 
 	OUT_PKT4(ring, REG_A6XX_SP_UNKNOWN_A0F8, 1);
-	OUT_RING(ring, 0x1);
+	OUT_RING(ring, fd6_ctx->magic.SP_UNKNOWN_A0F8);
 
 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
 	OUT_RING(ring, UNK_2C);
@@ -717,7 +718,7 @@ emit_binning_pass(struct fd_batch *batch)
 	OUT_WFI5(ring);
 
 	OUT_PKT4(ring, REG_A6XX_RB_CCU_CNTL, 1);
-	OUT_RING(ring, 0x7c400004);        /* RB_CCU_CNTL */
+	OUT_RING(ring, fd6_ctx->magic.RB_CCU_CNTL_gmem);
 }
 
 static void
@@ -771,10 +772,9 @@ fd6_emit_tile_init(struct fd_batch *batch)
 	OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
 	OUT_RING(ring, 0x0);
 
-	/* 0x10000000 for BYPASS.. 0x7c13c080 for GMEM: */
 	fd_wfi(batch, ring);
 	OUT_PKT4(ring, REG_A6XX_RB_CCU_CNTL, 1);
-	OUT_RING(ring, 0x7c400004);   /* RB_CCU_CNTL */
+	OUT_RING(ring, fd6_context(ctx)->magic.RB_CCU_CNTL_gmem);
 
 	emit_zs(ring, pfb->zsbuf, &ctx->gmem);
 	emit_mrt(ring, pfb, &ctx->gmem);
@@ -808,10 +808,10 @@ fd6_emit_tile_init(struct fd_batch *batch)
 		OUT_RING(ring, 0x0);
 
 		OUT_PKT4(ring, REG_A6XX_PC_UNKNOWN_9805, 1);
-		OUT_RING(ring, 0x1);
+		OUT_RING(ring, fd6_context(ctx)->magic.PC_UNKNOWN_9805);
 
 		OUT_PKT4(ring, REG_A6XX_SP_UNKNOWN_A0F8, 1);
-		OUT_RING(ring, 0x1);
+		OUT_RING(ring, fd6_context(ctx)->magic.SP_UNKNOWN_A0F8);
 
 		OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
 		OUT_RING(ring, 0x1);
@@ -1463,10 +1463,9 @@ fd6_emit_sysmem_prep(struct fd_batch *batch)
 	fd6_event_write(batch, ring, PC_CCU_INVALIDATE_COLOR, false);
 	fd6_cache_inv(batch, ring);
 
-	/* 0x10000000 for BYPASS.. 0x7c13c080 for GMEM: */
 	fd_wfi(batch, ring);
 	OUT_PKT4(ring, REG_A6XX_RB_CCU_CNTL, 1);
-	OUT_RING(ring, 0x10000000);   /* RB_CCU_CNTL */
+	OUT_RING(ring, fd6_context(batch->ctx)->magic.RB_CCU_CNTL_bypass);
 
 	/* enable stream-out, with sysmem there is only one pass: */
 	OUT_PKT4(ring, REG_A6XX_VPC_SO_OVERRIDE, 1);