freedreno/a4xx: better workaround for astc+srgb

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
setting up a second texture state and using that to sample alpha
separately.

This way, srgb->linear conversion happens in hw *prior* to
interpolation.

This fixes 546 dEQP tests: dEQP-GLES3.functional.texture.*astc*srgb*

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

5 files changed, 88 insertions, 10 deletions

diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_context.h b/src/gallium/drivers/freedreno/a4xx/fd4_context.h
index 8996de932b8..9467fc529bf 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 astc srgb workaround: */
+	uint16_t vastc_srgb, fastc_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..68e1f53e2dd 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->vastc_srgb != key->vastc_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->fastc_srgb != key->fastc_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->fastc_srgb || fd4_ctx->vastc_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,
+			.vastc_srgb = fd4_ctx->vastc_srgb,
+			.fastc_srgb = fd4_ctx->fastc_srgb,
 		},
 		.rasterflat = ctx->rasterizer->flatshade,
 		.sprite_coord_enable = ctx->rasterizer->sprite_coord_enable,
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
index ba5d48909fa..27614f07de5 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
@@ -123,7 +123,8 @@ fd4_emit_const_bo(struct fd_ringbuffer *ring, enum shader_t type, boolean write,
 
 static void
 emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
-		enum adreno_state_block sb, struct fd_texture_stateobj *tex)
+		enum adreno_state_block sb, struct fd_texture_stateobj *tex,
+		const struct ir3_shader_variant *v)
 {
 	static const uint32_t bcolor_reg[] = {
 			[SB_VERT_TEX] = REG_A4XX_TPL1_TP_VS_BORDER_COLOR_BASE_ADDR,
@@ -174,12 +175,14 @@ emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	if (tex->num_textures > 0) {
+		unsigned num_textures = tex->num_textures + v->astc_srgb.count;
+
 		/* emit texture state: */
-		OUT_PKT3(ring, CP_LOAD_STATE, 2 + (8 * tex->num_textures));
+		OUT_PKT3(ring, CP_LOAD_STATE, 2 + (8 * num_textures));
 		OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
 				CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
 				CP_LOAD_STATE_0_STATE_BLOCK(sb) |
-				CP_LOAD_STATE_0_NUM_UNIT(tex->num_textures));
+				CP_LOAD_STATE_0_NUM_UNIT(num_textures));
 		OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
 				CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
 		for (i = 0; i < tex->num_textures; i++) {
@@ -202,6 +205,34 @@ emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
 			OUT_RING(ring, 0x00000000);
 			OUT_RING(ring, 0x00000000);
 		}
+
+		for (i = 0; i < v->astc_srgb.count; i++) {
+			static const struct fd4_pipe_sampler_view dummy_view = {};
+			const struct fd4_pipe_sampler_view *view;
+			unsigned idx = v->astc_srgb.orig_idx[i];
+
+			view = tex->textures[idx] ?
+					fd4_pipe_sampler_view(tex->textures[idx]) :
+					&dummy_view;
+
+			debug_assert(view->texconst0 & A4XX_TEX_CONST_0_SRGB);
+
+			OUT_RING(ring, view->texconst0 & ~A4XX_TEX_CONST_0_SRGB);
+			OUT_RING(ring, view->texconst1);
+			OUT_RING(ring, view->texconst2);
+			OUT_RING(ring, view->texconst3);
+			if (view->base.texture) {
+				struct fd_resource *rsc = fd_resource(view->base.texture);
+				OUT_RELOC(ring, rsc->bo, view->offset, view->texconst4, 0);
+			} else {
+				OUT_RING(ring, 0x00000000);
+			}
+			OUT_RING(ring, 0x00000000);
+			OUT_RING(ring, 0x00000000);
+			OUT_RING(ring, 0x00000000);
+		}
+	} else {
+		debug_assert(v->astc_srgb.count == 0);
 	}
 
 	OUT_PKT0(ring, bcolor_reg[sb], 1);
@@ -681,14 +712,14 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	if (dirty & FD_DIRTY_VERTTEX) {
 		if (vp->has_samp)
-			emit_textures(ctx, ring, SB_VERT_TEX, &ctx->verttex);
+			emit_textures(ctx, ring, SB_VERT_TEX, &ctx->verttex, vp);
 		else
 			dirty &= ~FD_DIRTY_VERTTEX;
 	}
 
 	if (dirty & FD_DIRTY_FRAGTEX) {
 		if (fp->has_samp)
-			emit_textures(ctx, ring, SB_FRAG_TEX, &ctx->fragtex);
+			emit_textures(ctx, ring, SB_FRAG_TEX, &ctx->fragtex, fp);
 		else
 			dirty &= ~FD_DIRTY_FRAGTEX;
 	}
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_texture.c b/src/gallium/drivers/freedreno/a4xx/fd4_texture.c
index 38348580e21..6d9ecb7da24 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_astc_srgb_workaround(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 +240,11 @@ 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))
+	if (util_format_is_srgb(cso->format)) {
+		if (use_astc_srgb_workaround(pctx, cso->format))
+			so->astc_srgb = true;
 		so->texconst0 |= A4XX_TEX_CONST_0_SRGB;
+	}
 
 	if (cso->target == PIPE_BUFFER) {
 		unsigned elements = cso->u.buf.last_element -
@@ -296,11 +306,39 @@ 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 astc_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->astc_srgb)
+				astc_srgb |= (1 << i);
+		}
+	}
+
+	fd_set_sampler_views(pctx, shader, start, nr, views);
+
+	if (shader == PIPE_SHADER_FRAGMENT) {
+		fd4_ctx->fastc_srgb = astc_srgb;
+	} else if (shader == PIPE_SHADER_VERTEX) {
+		fd4_ctx->vastc_srgb = astc_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..21ceadd8c63 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 astc_srgb;
 };
 
 static inline struct fd4_pipe_sampler_view *