[Mesa-dev] [PATCH 2/2] radeonsi: eliminate PS color outputs when colormask kills them

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

---
 src/gallium/drivers/radeonsi/si_state.c         | 4 ++++
 src/gallium/drivers/radeonsi/si_state.h         | 1 +
 src/gallium/drivers/radeonsi/si_state_shaders.c | 1 +
 3 files changed, 6 insertions(+)

diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 41b08f8..5ee8bb9 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -434,20 +434,22 @@ static void *si_create_blend_state_mode(struct pipe_context *ctx,
 		       S_028B70_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
 		       S_028B70_ALPHA_TO_MASK_OFFSET0(2) |
 		       S_028B70_ALPHA_TO_MASK_OFFSET1(2) |
 		       S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
 		       S_028B70_ALPHA_TO_MASK_OFFSET3(2));
 
 	if (state->alpha_to_coverage)
 		blend->need_src_alpha_4bit |= 0xf;
 
 	blend->cb_target_mask = 0;
+	blend->cb_target_enabled_4bit = 0;
+
 	for (int i = 0; i < 8; i++) {
 		/* state->rt entries > 0 only written if independent blending */
 		const int j = state->independent_blend_enable ? i : 0;
 
 		unsigned eqRGB = state->rt[j].rgb_func;
 		unsigned srcRGB = state->rt[j].rgb_src_factor;
 		unsigned dstRGB = state->rt[j].rgb_dst_factor;
 		unsigned eqA = state->rt[j].alpha_func;
 		unsigned srcA = state->rt[j].alpha_src_factor;
 		unsigned dstA = state->rt[j].alpha_dst_factor;
@@ -475,20 +477,22 @@ static void *si_create_blend_state_mode(struct pipe_context *ctx,
 		if (blend->dual_src_blend &&
 		    (eqRGB == PIPE_BLEND_MIN || eqRGB == PIPE_BLEND_MAX ||
 		     eqA == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MAX)) {
 			assert(!"Unsupported equation for dual source blending");
 			si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
 			continue;
 		}
 
 		/* cb_render_state will disable unused ones */
 		blend->cb_target_mask |= (unsigned)state->rt[j].colormask << (4 * i);
+		if (state->rt[j].colormask)
+			blend->cb_target_enabled_4bit |= 0xf << (4 * i);
 
 		if (!state->rt[j].colormask || !state->rt[j].blend_enable) {
 			si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
 			continue;
 		}
 
 		/* Blending optimizations for RB+.
 		 * These transformations don't change the behavior.
 		 *
 		 * First, get rid of DST in the blend factors:
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 26c7b4c..7b7d96c 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -48,20 +48,21 @@ struct si_shader_selector;
 
 struct si_state_blend {
 	struct si_pm4_state	pm4;
 	uint32_t		cb_target_mask;
 	bool			alpha_to_coverage;
 	bool			alpha_to_one;
 	bool			dual_src_blend;
 	/* Set 0xf or 0x0 (4 bits) per render target if the following is
 	 * true. ANDed with spi_shader_col_format.
 	 */
+	unsigned		cb_target_enabled_4bit;
 	unsigned		blend_enable_4bit;
 	unsigned		need_src_alpha_4bit;
 };
 
 struct si_state_rasterizer {
 	struct si_pm4_state	pm4;
 	/* poly offset states for 16-bit, 24-bit, and 32-bit zbuffers */
 	struct si_pm4_state	*pm4_poly_offset;
 	unsigned		pa_sc_line_stipple;
 	unsigned		pa_cl_clip_cntl;
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 71d7987..061b3d2 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -1348,20 +1348,21 @@ static inline void si_shader_selector_key(struct pipe_context *ctx,
 			 */
 			key->part.ps.epilog.spi_shader_col_format =
 				(blend->blend_enable_4bit & blend->need_src_alpha_4bit &
 				 sctx->framebuffer.spi_shader_col_format_blend_alpha) |
 				(blend->blend_enable_4bit & ~blend->need_src_alpha_4bit &
 				 sctx->framebuffer.spi_shader_col_format_blend) |
 				(~blend->blend_enable_4bit & blend->need_src_alpha_4bit &
 				 sctx->framebuffer.spi_shader_col_format_alpha) |
 				(~blend->blend_enable_4bit & ~blend->need_src_alpha_4bit &
 				 sctx->framebuffer.spi_shader_col_format);
+			key->part.ps.epilog.spi_shader_col_format &= blend->cb_target_enabled_4bit;
 
 			/* The output for dual source blending should have
 			 * the same format as the first output.
 			 */
 			if (blend->dual_src_blend)
 				key->part.ps.epilog.spi_shader_col_format |=
 					(key->part.ps.epilog.spi_shader_col_format & 0xf) << 4;
 		} else
 			key->part.ps.epilog.spi_shader_col_format = sctx->framebuffer.spi_shader_col_format;
 
-- 
2.7.4