agx: Do not allow creating vec8

mem_access_bit_size needs to split up 64x4 into 2 loads. Fixes:

dEQP-VK.spirv_assembly.instruction.compute.64bit_compare.int64.comp_opiequal_vector

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24635>

diff --git a/src/asahi/compiler/agx_compile.c b/src/asahi/compiler/agx_compile.c
index c21c8b8..b1010de 100644 (file)
--- a/src/asahi/compiler/agx_compile.c
+++ b/src/asahi/compiler/agx_compile.c
@@ -2445,7 +2445,7 @@ mem_access_size_align_cb(nir_intrinsic_op intrin, uint8_t bytes,
       bit_size = 32;
 
    return (nir_mem_access_size_align){
-      .num_components = bytes / (bit_size / 8),
+      .num_components = MIN2(bytes / (bit_size / 8), 4),
       .bit_size = bit_size,
       .align = bit_size / 8,
    };