broadcom/vc5: Emit flat shade flags for varying components > 24.

This means that with no flatshading we'll emit the single-byte
ZERO_ALL_FLAT_SHADE_FLAGS, and otherwise emit a set of FLAT_SHADE_FLAGS to
get all the bits we need set.

There's a _SET enum in the packet we could use to possibly set entire
ranges of the bitfield without using another packet, but this at least
fixes the conformance failure.

3 files changed, 20 insertions, 7 deletions

diff --git a/src/broadcom/cle/v3d_packet_v33.xml b/src/broadcom/cle/v3d_packet_v33.xml
index 0402484dd71..6ec5332aeb7 100644
--- a/src/broadcom/cle/v3d_packet_v33.xml
+++ b/src/broadcom/cle/v3d_packet_v33.xml
@@ -92,6 +92,12 @@
     <value name="ANISOTROPIC_16_1" value="15"/>
   </enum>
 
+  <enum name="Flat Shade Action" prefix="V3D_FLAT_SHADE_ACTION">
+    <value name="unchanged" value="0"/>
+    <value name="zeroed" value="1"/>
+    <value name="set" value="2"/>
+  </enum>
+
   <packet code="0" name="Halt"/>
   <packet code="1" name="NOP"/>
   <packet code="4" name="Flush"/>
@@ -357,8 +363,8 @@
 
   <packet code="98" name="Flat Shade Flags">
     <field name="Flat Shade Flags for varyings V0*24" size="24" start="8" type="uint"/>
-    <field name="Action for Flat Shade Flags of higher numbered varyings" size="2" start="6" type="uint"/>
-    <field name="Action for Flat Shade Flags of lower numbered varyings" size="2" start="4" type="uint"/>
+    <field name="Action for Flat Shade Flags of higher numbered varyings" size="2" start="6" type="Flat Shade Action"/>
+    <field name="Action for Flat Shade Flags of lower numbered varyings" size="2" start="4" type="Flat Shade Action"/>
     <field name="Varying offset V0" size="4" start="0" type="uint"/>
   </packet>
 
diff --git a/src/broadcom/compiler/v3d_compiler.h b/src/broadcom/compiler/v3d_compiler.h
index 99d58e73ae8..85def2cb02c 100644
--- a/src/broadcom/compiler/v3d_compiler.h
+++ b/src/broadcom/compiler/v3d_compiler.h
@@ -421,7 +421,7 @@ struct v3d_compile {
          * flat-shaded.  This includes gl_FragColor flat-shading, which is
          * customized based on the shademodel_flat shader key.
          */
-        BITSET_WORD flat_shade_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
+        uint32_t flat_shade_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
 
         struct v3d_ubo_range *ubo_ranges;
         bool *ubo_range_used;
@@ -569,9 +569,12 @@ struct v3d_fs_prog_data {
 
         struct v3d_varying_slot input_slots[V3D_MAX_FS_INPUTS];
 
-        /* Bitmask for whether the corresponding input is flat-shaded.
+        /* Array of flat shade flags.
+         *
+         * Each entry is only 24 bits (high 8 bits 0), to match the hardware
+         * packet layout.
          */
-        BITSET_WORD flat_shade_flags[BITSET_WORDS(V3D_MAX_FS_INPUTS)];
+        uint32_t flat_shade_flags[((V3D_MAX_FS_INPUTS - 1) / 24) + 1];
 
         bool writes_z;
         bool discard;
diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c
index 2589c7f554d..4e78a477bd7 100644
--- a/src/broadcom/compiler/vir.c
+++ b/src/broadcom/compiler/vir.c
@@ -714,8 +714,12 @@ v3d_set_fs_prog_data_inputs(struct v3d_compile *c,
         memcpy(prog_data->input_slots, c->input_slots,
                c->num_inputs * sizeof(*c->input_slots));
 
-        memcpy(prog_data->flat_shade_flags, c->flat_shade_flags,
-               sizeof(c->flat_shade_flags));
+        STATIC_ASSERT(ARRAY_SIZE(prog_data->flat_shade_flags) >
+                      (V3D_MAX_FS_INPUTS - 1) / 24);
+        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);
+        }
 }
 
 uint64_t *v3d_compile_fs(const struct v3d_compiler *compiler,