freedreno/a3xx: add support to emulate GL_CLAMP

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

8 files changed, 129 insertions, 16 deletions

diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_context.h b/src/gallium/drivers/freedreno/a3xx/fd3_context.h
index 7d7663a28bc..48bbb47e9ca 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_context.h
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_context.h
@@ -76,6 +76,16 @@ struct fd3_context {
 
 	struct u_upload_mgr *border_color_uploader;
 	struct pipe_resource *border_color_buf;
+
+	/* bitmask of sampler which needs coords clamped for vertex
+	 * shader:
+	 */
+	unsigned vsaturate_s, vsaturate_t, vsaturate_r;
+
+	/* bitmask of sampler which needs coords clamped for frag
+	 * shader:
+	 */
+	unsigned fsaturate_s, fsaturate_t, fsaturate_r;
 };
 
 static INLINE struct fd3_context *
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_draw.c b/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
index 15d2ce4a943..f7a5fcafd9a 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
@@ -100,6 +100,7 @@ static void
 fd3_draw(struct fd_context *ctx, const struct pipe_draw_info *info)
 {
 	unsigned dirty = ctx->dirty;
+	struct fd3_context *fd3_ctx = fd3_context(ctx);
 	struct ir3_shader_key key = {
 			/* do binning pass first: */
 			.binning_pass = true,
@@ -108,7 +109,14 @@ fd3_draw(struct fd_context *ctx, const struct pipe_draw_info *info)
 			// TODO set .half_precision based on render target format,
 			// ie. float16 and smaller use half, float32 use full..
 			.half_precision = !!(fd_mesa_debug & FD_DBG_FRAGHALF),
+			.vsaturate_s = fd3_ctx->vsaturate_s,
+			.vsaturate_t = fd3_ctx->vsaturate_t,
+			.vsaturate_r = fd3_ctx->vsaturate_r,
+			.fsaturate_s = fd3_ctx->fsaturate_s,
+			.fsaturate_t = fd3_ctx->fsaturate_t,
+			.fsaturate_r = fd3_ctx->fsaturate_r,
 	};
+
 	draw_impl(ctx, info, ctx->binning_ring,
 			dirty & ~(FD_DIRTY_BLEND), key);
 	/* and now regular (non-binning) pass: */
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_texture.c b/src/gallium/drivers/freedreno/a3xx/fd3_texture.c
index 918dcc491f2..d70b39e2114 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_texture.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_texture.c
@@ -36,30 +36,31 @@
 #include "fd3_util.h"
 
 static enum a3xx_tex_clamp
-tex_clamp(unsigned wrap)
+tex_clamp(unsigned wrap, bool clamp_to_edge)
 {
-	/* hardware probably supports more, but we can't coax all the
-	 * wrap/clamp modes out of the GLESv2 blob driver.
-	 *
-	 * TODO once we have basics working, go back and just try
-	 * different values and see what happens
-	 */
+	/* Hardware does not support _CLAMP, but we emulate it: */
+	if (wrap == PIPE_TEX_WRAP_CLAMP) {
+		wrap = (clamp_to_edge) ?
+			PIPE_TEX_WRAP_CLAMP_TO_EDGE : PIPE_TEX_WRAP_CLAMP_TO_BORDER;
+	}
+
 	switch (wrap) {
 	case PIPE_TEX_WRAP_REPEAT:
 		return A3XX_TEX_REPEAT;
-	case PIPE_TEX_WRAP_CLAMP:
 	case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
 		return A3XX_TEX_CLAMP_TO_EDGE;
 	case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
 		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;
+	case PIPE_TEX_WRAP_MIRROR_CLAMP:
+	case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+		/* these two we could perhaps emulate, but we currently
+		 * just don't advertise PIPE_CAP_TEXTURE_MIRROR_CLAMP
+		 */
 	default:
 		DBG("invalid wrap: %u", wrap);
 		return 0;
@@ -86,6 +87,7 @@ fd3_sampler_state_create(struct pipe_context *pctx,
 {
 	struct fd3_sampler_stateobj *so = CALLOC_STRUCT(fd3_sampler_stateobj);
 	bool miplinear = false;
+	bool clamp_to_edge;
 
 	if (!so)
 		return NULL;
@@ -95,14 +97,29 @@ fd3_sampler_state_create(struct pipe_context *pctx,
 
 	so->base = *cso;
 
+	/*
+	 * For nearest filtering, _CLAMP means _CLAMP_TO_EDGE;  for linear
+	 * filtering, _CLAMP means _CLAMP_TO_BORDER while additionally
+	 * clamping the texture coordinates to [0.0, 1.0].
+	 *
+	 * The clamping will be taken care of in the shaders.  There are two
+	 * filters here, but let the minification one has a say.
+	 */
+	clamp_to_edge = (cso->min_img_filter == PIPE_TEX_FILTER_NEAREST);
+	if (!clamp_to_edge) {
+		so->saturate_s = (cso->wrap_s == PIPE_TEX_WRAP_CLAMP);
+		so->saturate_t = (cso->wrap_t == PIPE_TEX_WRAP_CLAMP);
+		so->saturate_r = (cso->wrap_r == PIPE_TEX_WRAP_CLAMP);
+	}
+
 	so->texsamp0 =
 			COND(!cso->normalized_coords, A3XX_TEX_SAMP_0_UNNORM_COORDS) |
 			COND(miplinear, A3XX_TEX_SAMP_0_MIPFILTER_LINEAR) |
 			A3XX_TEX_SAMP_0_XY_MAG(tex_filter(cso->mag_img_filter)) |
 			A3XX_TEX_SAMP_0_XY_MIN(tex_filter(cso->min_img_filter)) |
-			A3XX_TEX_SAMP_0_WRAP_S(tex_clamp(cso->wrap_s)) |
-			A3XX_TEX_SAMP_0_WRAP_T(tex_clamp(cso->wrap_t)) |
-			A3XX_TEX_SAMP_0_WRAP_R(tex_clamp(cso->wrap_r));
+			A3XX_TEX_SAMP_0_WRAP_S(tex_clamp(cso->wrap_s, clamp_to_edge)) |
+			A3XX_TEX_SAMP_0_WRAP_T(tex_clamp(cso->wrap_t, clamp_to_edge)) |
+			A3XX_TEX_SAMP_0_WRAP_R(tex_clamp(cso->wrap_r, clamp_to_edge));
 
 	if (cso->compare_mode)
 		so->texsamp0 |= A3XX_TEX_SAMP_0_COMPARE_FUNC(cso->compare_func); /* maps 1:1 */
@@ -124,7 +141,35 @@ fd3_sampler_states_bind(struct pipe_context *pctx,
 		unsigned shader, unsigned start,
 		unsigned nr, void **hwcso)
 {
+	struct fd_context *ctx = fd_context(pctx);
+	struct fd3_context *fd3_ctx = fd3_context(ctx);
+	unsigned saturate_s = 0, saturate_t = 0, saturate_r = 0;
+	unsigned i;
+
+	for (i = 0; i < nr; i++) {
+		if (hwcso[i]) {
+			struct fd3_sampler_stateobj *sampler =
+					fd3_sampler_stateobj(hwcso[i]);
+			if (sampler->saturate_s)
+				saturate_s |= (1 << i);
+			if (sampler->saturate_t)
+				saturate_t |= (1 << i);
+			if (sampler->saturate_r)
+				saturate_r |= (1 << i);
+		}
+	}
+
 	fd_sampler_states_bind(pctx, shader, start, nr, hwcso);
+
+	if (shader == PIPE_SHADER_FRAGMENT) {
+		fd3_ctx->fsaturate_s = saturate_s;
+		fd3_ctx->fsaturate_t = saturate_t;
+		fd3_ctx->fsaturate_r = saturate_r;
+	} else if (shader == PIPE_SHADER_VERTEX) {
+		fd3_ctx->vsaturate_s = saturate_s;
+		fd3_ctx->vsaturate_t = saturate_t;
+		fd3_ctx->vsaturate_r = saturate_r;
+	}
 }
 
 static enum a3xx_tex_type
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_texture.h b/src/gallium/drivers/freedreno/a3xx/fd3_texture.h
index a83f527366b..0434df37413 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_texture.h
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_texture.h
@@ -40,6 +40,7 @@
 struct fd3_sampler_stateobj {
 	struct pipe_sampler_state base;
 	uint32_t texsamp0, texsamp1;
+	bool saturate_s, saturate_t, saturate_r;
 };
 
 static INLINE struct fd3_sampler_stateobj *
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler.c
index c2d4942b66d..07b57b8d049 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_compiler.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler.c
@@ -135,7 +135,7 @@ compile_init(struct ir3_compile_context *ctx, struct ir3_shader_variant *so,
 {
 	unsigned ret;
 	struct tgsi_shader_info *info = &ctx->info;
-	const struct fd_lowering_config lconfig = {
+	struct fd_lowering_config lconfig = {
 			.color_two_side = so->key.color_two_side,
 			.lower_DST  = true,
 			.lower_XPD  = true,
@@ -153,6 +153,20 @@ compile_init(struct ir3_compile_context *ctx, struct ir3_shader_variant *so,
 			.lower_DP2A = true,
 	};
 
+	switch (so->type) {
+	case SHADER_FRAGMENT:
+	case SHADER_COMPUTE:
+		lconfig.saturate_s = so->key.fsaturate_s;
+		lconfig.saturate_t = so->key.fsaturate_t;
+		lconfig.saturate_r = so->key.fsaturate_r;
+		break;
+	case SHADER_VERTEX:
+		lconfig.saturate_s = so->key.vsaturate_s;
+		lconfig.saturate_t = so->key.vsaturate_t;
+		lconfig.saturate_r = so->key.vsaturate_r;
+		break;
+	}
+
 	ctx->tokens = fd_transform_lowering(&lconfig, tokens, &ctx->info);
 	ctx->free_tokens = !!ctx->tokens;
 	if (!ctx->tokens) {
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler_old.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler_old.c
index 1e1ca7ad813..4267feb351f 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_compiler_old.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler_old.c
@@ -125,7 +125,7 @@ compile_init(struct ir3_compile_context *ctx, struct ir3_shader_variant *so,
 {
 	unsigned ret, base = 0;
 	struct tgsi_shader_info *info = &ctx->info;
-	const struct fd_lowering_config lconfig = {
+	struct fd_lowering_config lconfig = {
 			.color_two_side = so->key.color_two_side,
 			.lower_DST  = true,
 			.lower_XPD  = true,
@@ -143,6 +143,20 @@ compile_init(struct ir3_compile_context *ctx, struct ir3_shader_variant *so,
 			.lower_DP2A = true,
 	};
 
+	switch (so->type) {
+	case SHADER_FRAGMENT:
+	case SHADER_COMPUTE:
+		lconfig.saturate_s = so->key.fsaturate_s;
+		lconfig.saturate_t = so->key.fsaturate_t;
+		lconfig.saturate_r = so->key.fsaturate_r;
+		break;
+	case SHADER_VERTEX:
+		lconfig.saturate_s = so->key.vsaturate_s;
+		lconfig.saturate_t = so->key.vsaturate_t;
+		lconfig.saturate_r = so->key.vsaturate_r;
+		break;
+	}
+
 	ctx->tokens = fd_transform_lowering(&lconfig, tokens, &ctx->info);
 	ctx->free_tokens = !!ctx->tokens;
 	if (!ctx->tokens) {
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.c b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
index 6d45597886b..ed7c639c930 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
@@ -185,11 +185,17 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key)
 	 */
 	if (shader->type == SHADER_FRAGMENT) {
 		key.binning_pass = false;
+		key.vsaturate_s = 0;
+		key.vsaturate_t = 0;
+		key.vsaturate_r = 0;
 	}
 	if (shader->type == SHADER_VERTEX) {
 		key.color_two_side = false;
 		key.half_precision = false;
 		key.alpha = false;
+		key.fsaturate_s = 0;
+		key.fsaturate_t = 0;
+		key.fsaturate_r = 0;
 	}
 
 	for (v = shader->variants; v; v = v->next)
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.h b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
index ea861649176..04a737ef19d 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
@@ -52,8 +52,23 @@ static inline uint16_t sem2idx(ir3_semantic sem)
 /* Configuration key used to identify a shader variant.. different
  * shader variants can be used to implement features not supported
  * in hw (two sided color), binning-pass vertex shader, etc.
+ *
+ * TODO since shader key is starting to get larger (than 32bit)
+ * we probably should pass it around by ptr rather than value more
+ * of the places.. but watch out in ir3_shader_variant() where the
+ * key gets normalized, we need to make a copy there.
  */
 struct ir3_shader_key {
+	/* bitmask of sampler which needs coords clamped for vertex
+	 * shader:
+	 */
+	unsigned vsaturate_s, vsaturate_t, vsaturate_r;
+
+	/* bitmask of sampler which needs coords clamped for frag
+	 * shader:
+	 */
+	unsigned fsaturate_s, fsaturate_t, fsaturate_r;
+
 	/*
 	 * Vertex shader variant parameters:
 	 */