Mesa (master): freedreno/ir3: Set up outputs for multi-slot varyings.

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 29f58cfbd07b419bca2cbe1e455232c7319444f4
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=29f58cfbd07b419bca2cbe1e455232c7319444f4

Author: Eric Anholt <eric at anholt.net>
Date:   Wed Apr 22 12:22:30 2020 -0700

freedreno/ir3: Set up outputs for multi-slot varyings.

Necessary to avoid compiler assertion failures in:

dEQP-GLES31.functional.program_interface_query.program_output.type.interface_blocks.out.named_block_explicit_location.struct.mat3x2

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4562>

---

 src/freedreno/ir3/ir3_compiler_nir.c | 45 ++++++++++++++++++++----------------
 1 file changed, 25 insertions(+), 20 deletions(-)

diff --git a/src/freedreno/ir3/ir3_compiler_nir.c b/src/freedreno/ir3/ir3_compiler_nir.c
index 812ce7f9489..bd65f91e780 100644
--- a/src/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/freedreno/ir3/ir3_compiler_nir.c
@@ -3130,7 +3130,8 @@ static void
 setup_output(struct ir3_context *ctx, nir_variable *out)
 {
 	struct ir3_shader_variant *so = ctx->so;
-	unsigned ncomp = glsl_get_components(out->type);
+	unsigned slots = glsl_count_vec4_slots(out->type, false, false);
+	unsigned ncomp = glsl_get_components(glsl_without_array(out->type));
 	unsigned n = out->data.driver_location;
 	unsigned frac = out->data.location_frac;
 	unsigned slot = out->data.location;
@@ -3192,30 +3193,34 @@ setup_output(struct ir3_context *ctx, nir_variable *out)
 		ir3_context_error(ctx, "unknown shader type: %d\n", ctx->so->type);
 	}
 
-	compile_assert(ctx, n < ARRAY_SIZE(so->outputs));
 
-	so->outputs[n].slot = slot;
-	so->outputs_count = MAX2(so->outputs_count, n + 1);
+	so->outputs_count = out->data.driver_location + slots;
+	compile_assert(ctx, so->outputs_count < ARRAY_SIZE(so->outputs));
 
-	for (int i = 0; i < ncomp; i++) {
-		unsigned idx = (n * 4) + i + frac;
-		compile_assert(ctx, idx < ctx->noutputs);
-		ctx->outputs[idx] = create_immed(ctx->block, fui(0.0));
-	}
+	for (int i = 0; i < slots; i++) {
+		int slot_base = n + i;
+		so->outputs[slot_base].slot = slot + i;
 
-	/* if varying packing doesn't happen, we could end up in a situation
-	 * with "holes" in the output, and since the per-generation code that
-	 * sets up varying linkage registers doesn't expect to have more than
-	 * one varying per vec4 slot, pad the holes.
-	 *
-	 * Note that this should probably generate a performance warning of
-	 * some sort.
-	 */
-	for (int i = 0; i < frac; i++) {
-		unsigned idx = (n * 4) + i;
-		if (!ctx->outputs[idx]) {
+		for (int i = 0; i < ncomp; i++) {
+			unsigned idx = (slot_base * 4) + i + frac;
+			compile_assert(ctx, idx < ctx->noutputs);
 			ctx->outputs[idx] = create_immed(ctx->block, fui(0.0));
 		}
+
+		/* if varying packing doesn't happen, we could end up in a situation
+		 * with "holes" in the output, and since the per-generation code that
+		 * sets up varying linkage registers doesn't expect to have more than
+		 * one varying per vec4 slot, pad the holes.
+		 *
+		 * Note that this should probably generate a performance warning of
+		 * some sort.
+		 */
+		for (int i = 0; i < frac; i++) {
+			unsigned idx = (slot_base * 4) + i;
+			if (!ctx->outputs[idx]) {
+				ctx->outputs[idx] = create_immed(ctx->block, fui(0.0));
+			}
+		}
 	}
 }