sna/gen4+: Implement an opacity shader

Avoid the cumbersome lookup through the alpha gradient texture and
simply multiply the incoming opacity value. The next step will be to
reduce the number of floats required per vertex.

Now that we have removed the primary user of the alpha solid cache, it
may be time to retire that as well.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

3 files changed, 112 insertions, 0 deletions

diff --git a/src/sna/brw/brw.h b/src/sna/brw/brw.h
index f0f3ac87..e5fa72f9 100644
--- a/src/sna/brw/brw.h
+++ b/src/sna/brw/brw.h
@@ -12,3 +12,6 @@ bool brw_wm_kernel__projective(struct brw_compile *p, int dispatch_width);
 bool brw_wm_kernel__projective_mask(struct brw_compile *p, int dispatch_width);
 bool brw_wm_kernel__projective_mask_ca(struct brw_compile *p, int dispatch_width);
 bool brw_wm_kernel__projective_mask_sa(struct brw_compile *p, int dispatch_width);
+
+bool brw_wm_kernel__affine_opacity(struct brw_compile *p, int dispatch_width);
+bool brw_wm_kernel__projective_opacity(struct brw_compile *p, int dispatch_width);
diff --git a/src/sna/brw/brw_test_gen7.c b/src/sna/brw/brw_test_gen7.c
index c3f0e231..085b25cc 100644
--- a/src/sna/brw/brw_test_gen7.c
+++ b/src/sna/brw/brw_test_gen7.c
@@ -167,6 +167,17 @@ static void gen7_ps_nomask_projective(void)
 	compare(ps_kernel_nomask_projective);
 }
 
+static void gen7_ps_opacity(void)
+{
+	uint32_t store[1024];
+	struct brw_compile p;
+
+	brw_compile_init(&p, GEN, store);
+	brw_wm_kernel__affine_opacity(&p, 16);
+
+	compare(ps_kernel_nomask_affine);
+}
+
 void brw_test_gen7(void)
 {
 	gen7_ps_nomask_affine();
@@ -175,4 +186,6 @@ void brw_test_gen7(void)
 	gen7_ps_masksa_affine();
 
 	gen7_ps_nomask_projective();
+
+	gen7_ps_opacity();
 }
diff --git a/src/sna/brw/brw_wm.c b/src/sna/brw/brw_wm.c
index f96881af..bd4003dd 100644
--- a/src/sna/brw/brw_wm.c
+++ b/src/sna/brw/brw_wm.c
@@ -323,6 +323,68 @@ done:
 	brw_fb_write(p, dw);
 }
 
+static void brw_wm_write__opacity(struct brw_compile *p, int dw,
+				  int src, int mask)
+{
+	int n;
+
+	if (dw == 8 && p->gen >= 60) {
+		brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+
+		brw_MUL(p,
+			brw_message_reg(2),
+			brw_vec8_grf(src+0, 0),
+			brw_vec1_grf(mask, 3));
+		brw_MUL(p,
+			brw_message_reg(3),
+			brw_vec8_grf(src+1, 0),
+			brw_vec1_grf(mask, 3));
+		brw_MUL(p,
+			brw_message_reg(4),
+			brw_vec8_grf(src+2, 0),
+			brw_vec1_grf(mask, 3));
+		brw_MUL(p,
+			brw_message_reg(5),
+			brw_vec8_grf(src+3, 0),
+			brw_vec1_grf(mask, 3));
+
+		goto done;
+	}
+
+	brw_set_compression_control(p, BRW_COMPRESSION_COMPRESSED);
+
+	for (n = 0; n < 4; n++) {
+		if (p->gen >= 60) {
+			brw_MUL(p,
+				brw_message_reg(2 + 2*n),
+				brw_vec8_grf(src + 2*n, 0),
+				brw_vec1_grf(mask, 3));
+		} else if (p->gen >= 45 && dw == 16) {
+			brw_MUL(p,
+				brw_message_reg(2 + n + BRW_MRF_COMPR4),
+				brw_vec8_grf(src + 2*n, 0),
+				brw_vec1_grf(mask, 3));
+		} else {
+			brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+			brw_MUL(p,
+				brw_message_reg(2 + n),
+				brw_vec8_grf(src + 2*n, 0),
+				brw_vec1_grf(mask, 3));
+
+			if (dw == 16) {
+				brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
+				brw_MUL(p,
+					brw_message_reg(2 + n + 4),
+					brw_vec8_grf(src + 2*n+1, 0),
+					brw_vec1_grf(mask, 3));
+			}
+		}
+	}
+
+done:
+	brw_fb_write(p, dw);
+}
+
 static void brw_wm_write__mask_ca(struct brw_compile *p, int dw,
 				  int src, int mask)
 {
@@ -597,3 +659,37 @@ brw_wm_kernel__projective_mask_sa(struct brw_compile *p, int dispatch)
 
 	return true;
 }
+
+bool
+brw_wm_kernel__affine_opacity(struct brw_compile *p, int dispatch)
+{
+	int src, mask;
+
+	if (p->gen < 60) {
+		brw_wm_xy(p, dispatch);
+		mask = 4;
+	} else
+		mask = dispatch == 16 ? 8 : 6;
+
+	src = brw_wm_affine(p, dispatch, 0, 1, 12);
+	brw_wm_write__opacity(p, dispatch, src, mask);
+
+	return true;
+}
+
+bool
+brw_wm_kernel__projective_opacity(struct brw_compile *p, int dispatch)
+{
+	int src, mask;
+
+	if (p->gen < 60) {
+		brw_wm_xy(p, dispatch);
+		mask = 4;
+	} else
+		mask = dispatch == 16 ? 8 : 6;
+
+	src = brw_wm_projective(p, dispatch, 0, 1, 12);
+	brw_wm_write__opacity(p, dispatch, src, mask);
+
+	return true;
+}