Mesa (master): freedreno/ir3: handle scalarized varying inputs

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 4c060235a2e52db1d777ca193de57a447dbd6ec5
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=4c060235a2e52db1d777ca193de57a447dbd6ec5

Author: Rob Clark <robdclark at chromium.org>
Date:   Mon Oct 21 11:15:53 2019 -0700

freedreno/ir3: handle scalarized varying inputs

If the load_interpolated_input is scalarized, we would be too
conservative about deciding the tex instruction wasn't a candidate to
pre-fetch:

	vec1 32 ssa_0 = load_const (0x00000000 /* 0.000000 */)
	vec2 32 ssa_1 = intrinsic load_barycentric_pixel () (0) /* interp_mode=0 */
	vec1 32 ssa_2 = intrinsic load_interpolated_input (ssa_1, ssa_0) (0, 0) /* base=0 */ /* component=0 */	/* packed:v_uv,v_uv1 */
	vec1 32 ssa_3 = intrinsic load_interpolated_input (ssa_1, ssa_0) (0, 1) /* base=0 */ /* component=1 */	/* packed:v_uv,v_uv1 */
	vec2 32 ssa_8 = vec2 ssa_2, ssa_3
	vec4 32 ssa_9 = tex ssa_8 (coord), 0 (texture), 0 (sampler)

Really we don't care that the texcoord components come from different
load_interpolated_input instructions, just that they have consecutive
varying offsets.

Reported-by: Eduardo Lima Mitev <elima at igalia.com>
Reviewed-by: Eduardo Lima Mitev <elima at igalia.com>
Signed-off-by: Rob Clark <robdclark at chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg at google.com>

---

 src/freedreno/ir3/ir3_nir_lower_tex_prefetch.c | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)

diff --git a/src/freedreno/ir3/ir3_nir_lower_tex_prefetch.c b/src/freedreno/ir3/ir3_nir_lower_tex_prefetch.c
index 1a6fbb4ba5d..2faeb9b4169 100644
--- a/src/freedreno/ir3/ir3_nir_lower_tex_prefetch.c
+++ b/src/freedreno/ir3/ir3_nir_lower_tex_prefetch.c
@@ -45,22 +45,25 @@ coord_offset(nir_ssa_def *ssa)
 		if (alu->op != nir_op_vec2)
 			return -1;
 
-		for (int i = 0; i < 2; i++) {
+		if (!alu->src[0].src.is_ssa)
+			return -1;
+
+		int base_offset = coord_offset(alu->src[0].src.ssa) +
+				alu->src[0].swizzle[0];
+
+		/* NOTE it might be possible to support more than 2D? */
+		for (int i = 1; i < 2; i++) {
 			if (!alu->src[i].src.is_ssa)
 				return -1;
 
-			if (alu->src[i].swizzle[0] != (alu->src[0].swizzle[0] + i))
-				return -1;
+			int nth_offset = coord_offset(alu->src[i].src.ssa) +
+					alu->src[i].swizzle[0];
 
-			if (alu->src[i].src.ssa != alu->src[0].src.ssa)
+			if (nth_offset != (base_offset + i))
 				return -1;
 		}
 
-		int off = coord_offset(alu->src[0].src.ssa);
-		if (off < 0)
-			return -1;
-
-		return off + alu->src[0].swizzle[0];
+		return base_offset;
 	}
 
 	if (parent_instr->type != nir_instr_type_intrinsic)