broadcom/vc5: Add support for centroid varyings.

It would be nice to share the flags packet emit logic with flat shade
flags, but I couldn't come up with a good way while still using our pack
macros.  We need to refactor this to shader record setup at compile time,
anyway.

Fixes ext_framebuffer_multisample-interpolation * centroid-*

3 files changed, 44 insertions, 0 deletions

diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index d6c2d1902b6..ec8f22321f3 100644
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -436,6 +436,7 @@ emit_fragment_varying(struct v3d_compile *c, nir_variable *var,
                 /* FALLTHROUGH */
         case INTERP_MODE_SMOOTH:
                 if (var->data.centroid) {
+                        BITSET_SET(c->centroid_flags, i);
                         return vir_FADD(c, vir_FMUL(c, vary,
                                                     c->payload_w_centroid), r5);
                 } else {
@@ -1985,6 +1986,36 @@ vir_emit_last_thrsw(struct v3d_compile *c)
                 c->last_thrsw->is_last_thrsw = true;
 }
 
+/* There's a flag in the shader for "centroid W used in addition to center W",
+ * so we need to walk the program after VIR optimization to see if both are
+ * used.
+ */
+static void
+vir_check_payload_w(struct v3d_compile *c)
+{
+        if (c->s->info.stage != MESA_SHADER_FRAGMENT)
+                return;
+
+        bool any_centroid = false;
+        for (int i = 0; i < ARRAY_SIZE(c->centroid_flags); i++) {
+                if (c->centroid_flags[i])
+                        any_centroid = true;
+        }
+        if (!any_centroid)
+                return;
+
+        vir_for_each_inst_inorder(inst, c) {
+                for (int i = 0; i < vir_get_nsrc(inst); i++) {
+                        if (inst->src[i].file == QFILE_REG &&
+                            inst->src[i].index == 0) {
+                                c->uses_centroid_and_center_w = true;
+                                return;
+                        }
+                }
+        }
+
+}
+
 void
 v3d_nir_to_vir(struct v3d_compile *c)
 {
@@ -2024,6 +2055,8 @@ v3d_nir_to_vir(struct v3d_compile *c)
         vir_optimize(c);
         vir_lower_uniforms(c);
 
+        vir_check_payload_w(c);
+
         /* XXX: vir_schedule_instructions(c); */
 
         if (V3D_DEBUG & (V3D_DEBUG_VIR |
diff --git a/src/broadcom/compiler/v3d_compiler.h b/src/broadcom/compiler/v3d_compiler.h
index e89ea7be217..4dba23c067e 100644
--- a/src/broadcom/compiler/v3d_compiler.h
+++ b/src/broadcom/compiler/v3d_compiler.h
@@ -478,6 +478,10 @@ struct v3d_compile {
          */
         uint32_t flat_shade_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
 
+        uint32_t centroid_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
+
+        bool uses_centroid_and_center_w;
+
         struct v3d_ubo_range *ubo_ranges;
         bool *ubo_range_used;
         uint32_t ubo_ranges_array_size;
@@ -657,8 +661,11 @@ struct v3d_fs_prog_data {
          */
         uint32_t flat_shade_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
 
+        uint32_t centroid_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
+
         bool writes_z;
         bool discard;
+        bool uses_centroid_and_center_w;
 };
 
 /* Special nir_load_input intrinsic index for loading the current TLB
diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c
index 93990ee806f..0de5335d12a 100644
--- a/src/broadcom/compiler/vir.c
+++ b/src/broadcom/compiler/vir.c
@@ -758,6 +758,9 @@ v3d_set_fs_prog_data_inputs(struct v3d_compile *c,
         for (int i = 0; i < V3D_MAX_FS_INPUTS; i++) {
                 if (BITSET_TEST(c->flat_shade_flags, i))
                         prog_data->flat_shade_flags[i / 24] |= 1 << (i % 24);
+
+                if (BITSET_TEST(c->centroid_flags, i))
+                        prog_data->centroid_flags[i / 24] |= 1 << (i % 24);
         }
 }
 
@@ -838,6 +841,7 @@ uint64_t *v3d_compile_fs(const struct v3d_compiler *compiler,
         prog_data->writes_z = (c->s->info.outputs_written &
                                (1 << FRAG_RESULT_DEPTH));
         prog_data->discard = c->s->info.fs.uses_discard;
+        prog_data->uses_centroid_and_center_w = c->uses_centroid_and_center_w;
 
         return v3d_return_qpu_insts(c, final_assembly_size);
 }