pan/mdg: Mask spills from texture write
authorAlyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Fri, 31 Jul 2020 14:31:50 +0000 (10:31 -0400)
committerAlyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Mon, 3 Aug 2020 13:43:37 +0000 (09:43 -0400)
This prevents RA failures the results of reading multiple textures that
require less than 4 channels, as seen in a number of GL 3 WebRender shaders.

Closes: #3342
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reported-by: Icecream95 <ixn@keemail.me>
Tested-by: Icecream95 <ixn@keemail.me>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6144>

src/panfrost/midgard/midgard_ra.c

index 1b8c8c3d2de79ee32052324025f72f466ca8ba99..b06eb97ab8ec77f96e3126fc9bb0416a8fb9ca1a 100644 (file)
@@ -855,9 +855,17 @@ mir_spill_register(
 
                         midgard_instruction st;
 
+                        /* Note: it's important to match the mask of the spill
+                         * with the mask of the instruction whose destination
+                         * we're spilling, or otherwise we'll read invalid
+                         * components and can fail RA in a subsequent iteration
+                         */
+
                         if (is_special_w) {
                                 st = v_mov(spill_node, spill_slot);
                                 st.no_spill |= (1 << spill_class);
+                                st.mask = ins->mask;
+                                st.dest_type = st.src_types[0] = ins->dest_type;
                         } else {
                                 ins->dest = spill_index++;
                                 ins->no_spill |= (1 << spill_class);