freedreno/a4xx: lower srgb in shader for astc textures

This *seems* like a hw bug, and maybe only applies to certain a4xx
variants/revisions.  But setting the SRGB bit in sampler view state
(texconst0) causes invalid alpha for ASTC textures.  Work around this
by doing the srgb->linear conversion in the shader instead.

This fixes 392 dEQP tests: dEQP-GLES3.functional.texture.*astc*srgb*

(The remaining fails seem to be a bug w/ ASTC + linear filtering, also
possibly a420.0 specific.)

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

7 files changed, 62 insertions, 6 deletions

diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_context.h b/src/gallium/drivers/freedreno/a4xx/fd4_context.h
index 8996de932b8..a08c3c3119b 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_context.h
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_context.h
@@ -85,6 +85,9 @@ struct fd4_context {
 	 */
 	uint16_t fsaturate_s, fsaturate_t, fsaturate_r;
 
+	/* bitmask of samplers which need srgb lowering in vertex/frag shader: */
+	uint16_t vlower_srgb, flower_srgb;
+
 	/* some state changes require a different shader variant.  Keep
 	 * track of this so we know when we need to re-emit shader state
 	 * due to variant change.  See fixup_shader_state()
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_draw.c b/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
index e874d223187..a0866261fcb 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
@@ -93,12 +93,14 @@ fixup_shader_state(struct fd_context *ctx, struct ir3_shader_key *key)
 		if (last_key->has_per_samp || key->has_per_samp) {
 			if ((last_key->vsaturate_s != key->vsaturate_s) ||
 					(last_key->vsaturate_t != key->vsaturate_t) ||
-					(last_key->vsaturate_r != key->vsaturate_r))
+					(last_key->vsaturate_r != key->vsaturate_r) ||
+					(last_key->vlower_srgb != key->vlower_srgb))
 				ctx->prog.dirty |= FD_SHADER_DIRTY_VP;
 
 			if ((last_key->fsaturate_s != key->fsaturate_s) ||
 					(last_key->fsaturate_t != key->fsaturate_t) ||
-					(last_key->fsaturate_r != key->fsaturate_r))
+					(last_key->fsaturate_r != key->fsaturate_r) ||
+					(last_key->flower_srgb != key->flower_srgb))
 				ctx->prog.dirty |= FD_SHADER_DIRTY_FP;
 		}
 
@@ -132,13 +134,16 @@ fd4_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info)
 			// ie. float16 and smaller use half, float32 use full..
 			.half_precision = !!(fd_mesa_debug & FD_DBG_FRAGHALF),
 			.ucp_enables = ctx->rasterizer->clip_plane_enable,
-			.has_per_samp = (fd4_ctx->fsaturate || fd4_ctx->vsaturate),
+			.has_per_samp = (fd4_ctx->fsaturate || fd4_ctx->vsaturate ||
+					fd4_ctx->flower_srgb || fd4_ctx->vlower_srgb),
 			.vsaturate_s = fd4_ctx->vsaturate_s,
 			.vsaturate_t = fd4_ctx->vsaturate_t,
 			.vsaturate_r = fd4_ctx->vsaturate_r,
 			.fsaturate_s = fd4_ctx->fsaturate_s,
 			.fsaturate_t = fd4_ctx->fsaturate_t,
 			.fsaturate_r = fd4_ctx->fsaturate_r,
+			.vlower_srgb = fd4_ctx->vlower_srgb,
+			.flower_srgb = fd4_ctx->flower_srgb,
 		},
 		.rasterflat = ctx->rasterizer->flatshade,
 		.sprite_coord_enable = ctx->rasterizer->sprite_coord_enable,
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_texture.c b/src/gallium/drivers/freedreno/a4xx/fd4_texture.c
index 38348580e21..eee7a39adfd 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_texture.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_texture.c
@@ -209,6 +209,13 @@ tex_type(unsigned target)
 	}
 }
 
+static bool
+use_srgb_lowering(struct pipe_context *pctx, enum pipe_format format)
+{
+	return (fd_screen(pctx->screen)->gpu_id == 420) &&
+		(util_format_description(format)->layout == UTIL_FORMAT_LAYOUT_ASTC);
+}
+
 static struct pipe_sampler_view *
 fd4_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 		const struct pipe_sampler_view *cso)
@@ -233,8 +240,12 @@ fd4_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 		fd4_tex_swiz(cso->format, cso->swizzle_r, cso->swizzle_g,
 				cso->swizzle_b, cso->swizzle_a);
 
-	if (util_format_is_srgb(cso->format))
-		so->texconst0 |= A4XX_TEX_CONST_0_SRGB;
+	if (util_format_is_srgb(cso->format)) {
+		if (use_srgb_lowering(pctx, cso->format))
+			so->lower_srgb = true;
+		else
+			so->texconst0 |= A4XX_TEX_CONST_0_SRGB;
+	}
 
 	if (cso->target == PIPE_BUFFER) {
 		unsigned elements = cso->u.buf.last_element -
@@ -296,11 +307,39 @@ fd4_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 	return &so->base;
 }
 
+static void
+fd4_set_sampler_views(struct pipe_context *pctx, unsigned shader,
+		unsigned start, unsigned nr,
+		struct pipe_sampler_view **views)
+{
+	struct fd_context *ctx = fd_context(pctx);
+	struct fd4_context *fd4_ctx = fd4_context(ctx);
+	uint16_t lower_srgb = 0;
+	unsigned i;
+
+	for (i = 0; i < nr; i++) {
+		if (views[i]) {
+			struct fd4_pipe_sampler_view *view =
+					fd4_pipe_sampler_view(views[i]);
+			if (view->lower_srgb)
+				lower_srgb |= (1 << i);
+		}
+	}
+
+	fd_set_sampler_views(pctx, shader, start, nr, views);
+
+	if (shader == PIPE_SHADER_FRAGMENT) {
+		fd4_ctx->flower_srgb = lower_srgb;
+	} else if (shader == PIPE_SHADER_VERTEX) {
+		fd4_ctx->vlower_srgb = lower_srgb;
+	}
+}
+
 void
 fd4_texture_init(struct pipe_context *pctx)
 {
 	pctx->create_sampler_state = fd4_sampler_state_create;
 	pctx->bind_sampler_states = fd4_sampler_states_bind;
 	pctx->create_sampler_view = fd4_sampler_view_create;
-	pctx->set_sampler_views = fd_set_sampler_views;
+	pctx->set_sampler_views = fd4_set_sampler_views;
 }
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_texture.h b/src/gallium/drivers/freedreno/a4xx/fd4_texture.h
index 6ca34ade60d..af2140db504 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_texture.h
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_texture.h
@@ -53,6 +53,7 @@ struct fd4_pipe_sampler_view {
 	struct pipe_sampler_view base;
 	uint32_t texconst0, texconst1, texconst2, texconst3, texconst4;
 	uint32_t offset;
+	bool lower_srgb;
 };
 
 static inline struct fd4_pipe_sampler_view *
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_nir.c b/src/gallium/drivers/freedreno/ir3/ir3_nir.c
index 897b3b963be..7836789de8f 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_nir.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_nir.c
@@ -58,6 +58,7 @@ ir3_key_lowers_nir(const struct ir3_shader_key *key)
 {
 	return key->fsaturate_s | key->fsaturate_t | key->fsaturate_r |
 			key->vsaturate_s | key->vsaturate_t | key->vsaturate_r |
+			key->vlower_srgb | key->flower_srgb |
 			key->ucp_enables | key->color_two_side;
 }
 
@@ -85,11 +86,13 @@ ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s,
 			tex_options.saturate_s = key->fsaturate_s;
 			tex_options.saturate_t = key->fsaturate_t;
 			tex_options.saturate_r = key->fsaturate_r;
+			tex_options.lower_srgb = key->flower_srgb;
 			break;
 		case SHADER_VERTEX:
 			tex_options.saturate_s = key->vsaturate_s;
 			tex_options.saturate_t = key->vsaturate_t;
 			tex_options.saturate_r = key->vsaturate_r;
+			tex_options.lower_srgb = key->vlower_srgb;
 			break;
 		}
 	}
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.c b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
index c05b52e7a5e..3d38088510a 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
@@ -223,6 +223,7 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key)
 			key.vsaturate_s = 0;
 			key.vsaturate_t = 0;
 			key.vsaturate_r = 0;
+			key.vlower_srgb = 0;
 		}
 		break;
 	case SHADER_VERTEX:
@@ -233,6 +234,7 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key)
 			key.fsaturate_s = 0;
 			key.fsaturate_t = 0;
 			key.fsaturate_r = 0;
+			key.flower_srgb = 0;
 		}
 		break;
 	}
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.h b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
index c89dc29ff08..82cde51e635 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
@@ -104,6 +104,9 @@ struct ir3_shader_key {
 	 * shader:
 	 */
 	uint16_t fsaturate_s, fsaturate_t, fsaturate_r;
+
+	/* bitmask of samplers which need srgb->linear lowering: */
+	uint16_t vlower_srgb, flower_srgb;
 };
 
 static inline bool