freedreno/a3xx: initial texture border-color

Still some open questions.. and at any rate, no additional piglit passes
due to various wrap modes that we need to emulate in at least some
cases :-(

But it does fix some mystery page-faults.. So add some comments in the
code where there are things that we need to emulate or do more r/e, and
push as-is.

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

6 files changed, 65 insertions, 15 deletions

diff --git a/src/gallium/drivers/freedreno/a2xx/fd2_draw.c b/src/gallium/drivers/freedreno/a2xx/fd2_draw.c
index 77fbf1af224..bc7960a28a5 100644
--- a/src/gallium/drivers/freedreno/a2xx/fd2_draw.c
+++ b/src/gallium/drivers/freedreno/a2xx/fd2_draw.c
@@ -30,7 +30,6 @@
 #include "util/u_string.h"
 #include "util/u_memory.h"
 #include "util/u_prim.h"
-#include "util/u_pack_color.h"
 
 #include "freedreno_state.h"
 #include "freedreno_resource.h"
@@ -118,14 +117,6 @@ fd2_draw(struct fd_context *ctx, const struct pipe_draw_info *info)
 }
 
 
-static uint32_t
-pack_rgba(enum pipe_format format, const float *rgba)
-{
-	union util_color uc;
-	util_pack_color(rgba, format, &uc);
-	return uc.ui[0];
-}
-
 static void
 fd2_clear(struct fd_context *ctx, unsigned buffers,
 		const union pipe_color_union *color, double depth, unsigned stencil)
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_context.c b/src/gallium/drivers/freedreno/a3xx/fd3_context.c
index 847414ac082..f8f412e1a1b 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_context.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_context.c
@@ -137,5 +137,8 @@ fd3_context_create(struct pipe_screen *pscreen, void *priv)
 
 	fd3_query_context_init(pctx);
 
+	fd3_ctx->border_color_uploader = u_upload_create(pctx, 4096,
+			2 * PIPE_MAX_SAMPLERS * BORDERCOLOR_SIZE, 0);
+
 	return pctx;
 }
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_context.h b/src/gallium/drivers/freedreno/a3xx/fd3_context.h
index 26c8cc762b9..7d7663a28bc 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_context.h
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_context.h
@@ -29,6 +29,8 @@
 #ifndef FD3_CONTEXT_H_
 #define FD3_CONTEXT_H_
 
+#include "util/u_upload_mgr.h"
+
 #include "freedreno_drmif.h"
 
 #include "freedreno_context.h"
@@ -56,6 +58,24 @@ struct fd3_context {
 	/* vertex buf used for mem->gmem tex coords:
 	 */
 	struct pipe_resource *blit_texcoord_vbuf;
+
+	/*
+	 * Border color layout *appears* to be as arrays of 0x40 byte
+	 * elements, with frag shader elements starting at (16 x 0x40).
+	 * But at some point I should probably experiment more with
+	 * samplers in vertex shaders to be sure.  Unclear about why
+	 * there is this offset when there are separate VS and FS base
+	 * addr regs.
+	 *
+	 * The first 8 bytes of each entry are the requested border
+	 * color in fp16.  Unclear about the rest.. could be used for
+	 * other formats, or could simply be for aligning the pitch
+	 * to 32 pixels.
+	 */
+#define BORDERCOLOR_SIZE 0x40
+
+	struct u_upload_mgr *border_color_uploader;
+	struct pipe_resource *border_color_buf;
 };
 
 static INLINE struct fd3_context *
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
index 62e02bc6b2d..73a28dda5e6 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
@@ -152,9 +152,8 @@ emit_constants(struct fd_ringbuffer *ring,
 #define BASETABLE_SZ    A3XX_MAX_MIP_LEVELS
 
 static void
-emit_textures(struct fd_ringbuffer *ring,
-		enum adreno_state_block sb,
-		struct fd_texture_stateobj *tex)
+emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
+		enum adreno_state_block sb, struct fd_texture_stateobj *tex)
 {
 	static const unsigned tex_off[] = {
 			[SB_VERT_TEX] = VERT_TEX_OFF,
@@ -164,7 +163,18 @@ emit_textures(struct fd_ringbuffer *ring,
 			[SB_VERT_TEX] = SB_VERT_MIPADDR,
 			[SB_FRAG_TEX] = SB_FRAG_MIPADDR,
 	};
-	unsigned i, j;
+	static const uint32_t bcolor_reg[] = {
+			[SB_VERT_TEX] = REG_A3XX_TPL1_TP_VS_BORDER_COLOR_BASE_ADDR,
+			[SB_FRAG_TEX] = REG_A3XX_TPL1_TP_FS_BORDER_COLOR_BASE_ADDR,
+	};
+	struct fd3_context *fd3_ctx = fd3_context(ctx);
+	unsigned i, j, off;
+	void *ptr;
+
+	u_upload_alloc(fd3_ctx->border_color_uploader,
+			0, 2 * PIPE_MAX_SAMPLERS * BORDERCOLOR_SIZE, &off,
+			&fd3_ctx->border_color_buf,
+			&ptr);
 
 	if (tex->num_samplers > 0) {
 		/* output sampler state: */
@@ -180,6 +190,16 @@ emit_textures(struct fd_ringbuffer *ring,
 			const struct fd3_sampler_stateobj *sampler = tex->samplers[i] ?
 					fd3_sampler_stateobj(tex->samplers[i]) :
 					&dummy_sampler;
+			uint16_t *bcolor = (uint16_t *)((uint8_t *)ptr +
+					(BORDERCOLOR_SIZE * tex_off[sb]) +
+					(BORDERCOLOR_SIZE * i));
+
+			/* TODO not quite sure if bcolor is pre or post swizzle: */
+			for (j = 0; j < 4; j++) {
+				bcolor[j] =
+					util_float_to_half(sampler->base.border_color.f[j]);
+			}
+
 			OUT_RING(ring, sampler->texsamp0);
 			OUT_RING(ring, sampler->texsamp1);
 		}
@@ -237,6 +257,11 @@ emit_textures(struct fd_ringbuffer *ring,
 			}
 		}
 	}
+
+	OUT_PKT0(ring, bcolor_reg[sb], 1);
+	OUT_RELOC(ring, fd_resource(fd3_ctx->border_color_buf)->bo, off, 0, 0);
+
+	u_upload_unmap(fd3_ctx->border_color_uploader);
 }
 
 /* emit texture state for mem->gmem restore operation.. eventually it would
@@ -553,14 +578,14 @@ fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	if (dirty & FD_DIRTY_VERTTEX) {
 		if (vp->has_samp)
-			emit_textures(ring, SB_VERT_TEX, &ctx->verttex);
+			emit_textures(ctx, ring, SB_VERT_TEX, &ctx->verttex);
 		else
 			dirty &= ~FD_DIRTY_VERTTEX;
 	}
 
 	if (dirty & FD_DIRTY_FRAGTEX) {
 		if (fp->has_samp)
-			emit_textures(ring, SB_FRAG_TEX, &ctx->fragtex);
+			emit_textures(ctx, ring, SB_FRAG_TEX, &ctx->fragtex);
 		else
 			dirty &= ~FD_DIRTY_FRAGTEX;
 	}
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_texture.c b/src/gallium/drivers/freedreno/a3xx/fd3_texture.c
index 36a877dcd92..8308674178d 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_texture.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_texture.c
@@ -54,7 +54,9 @@ tex_clamp(unsigned wrap)
 		return A3XX_TEX_CLAMP_TO_BORDER;
 	case PIPE_TEX_WRAP_MIRROR_CLAMP:
 	case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+		/* these two we should emulate! */
 	case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+		/* only works for PoT.. need to emulate otherwise! */
 		return A3XX_TEX_MIRROR_CLAMP;
 	case PIPE_TEX_WRAP_MIRROR_REPEAT:
 		return A3XX_TEX_MIRROR_REPEAT;
diff --git a/src/gallium/drivers/freedreno/freedreno_util.h b/src/gallium/drivers/freedreno/freedreno_util.h
index d676f80d71b..a762c8f0575 100644
--- a/src/gallium/drivers/freedreno/freedreno_util.h
+++ b/src/gallium/drivers/freedreno/freedreno_util.h
@@ -38,6 +38,7 @@
 #include "util/u_math.h"
 #include "util/u_half.h"
 #include "util/u_dynarray.h"
+#include "util/u_pack_color.h"
 
 #include "adreno_common.xml.h"
 #include "adreno_pm4.xml.h"
@@ -250,4 +251,12 @@ static inline uint32_t env2u(const char *envvar)
 	return 0;
 }
 
+static inline uint32_t
+pack_rgba(enum pipe_format format, const float *rgba)
+{
+	union util_color uc;
+	util_pack_color(rgba, format, &uc);
+	return uc.ui[0];
+}
+
 #endif /* FREEDRENO_UTIL_H_ */