nir: reject unsupported component counts from all vectorize callbacks

If you allow an unsupported component count in the callback for loads,
nir_opt_load_store_vectorize will align num_components to the next supported
vector size, essentially overfetching.

This changes all callbacks to reject it. AMD will enable it in a later commit.

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29398>

src/broadcom/compiler/nir_to_vir.c

@@ -2089,7 +2089,7 @@ mem_vectorize_callback(unsigned align_mul, unsigned align_offset,
                        nir_intrinsic_instr *high,
                        void *data)
 {
-        if (hole_size)
+        if (hole_size || !nir_num_components_valid(num_components))
                 return false;
 
         /* TMU general access only supports 32-bit vectors */

src/compiler/nir/tests/load_store_vectorizer_tests.cpp

@@ -340,7 +340,7 @@ bool nir_load_store_vectorize_test::mem_vectorize_callback(
    nir_intrinsic_instr *low, nir_intrinsic_instr *high,
    void *data)
 {
-   if (hole_size)
+   if (hole_size || !nir_num_components_valid(num_components))
       return false;
 
    /* Calculate a simple alignment, like how nir_intrinsic_align() does. */

src/freedreno/ir3/ir3_nir.c

@@ -113,7 +113,7 @@ ir3_nir_should_vectorize_mem(unsigned align_mul, unsigned align_offset,
                              unsigned hole_size, nir_intrinsic_instr *low,
                              nir_intrinsic_instr *high, void *data)
 {
-   if (hole_size)
+   if (hole_size || !nir_num_components_valid(num_components))
       return false;
 
    struct ir3_compiler *compiler = data;

src/gallium/auxiliary/nir/nir_to_tgsi.c

@@ -3273,7 +3273,7 @@ ntt_should_vectorize_io(unsigned align, unsigned bit_size,
                         nir_intrinsic_instr *low, nir_intrinsic_instr *high,
                         void *data)
 {
-   if (bit_size != 32 || hole_size)
+   if (bit_size != 32 || hole_size || !nir_num_components_valid(num_components))
       return false;
 
    /* Our offset alignment should aways be at least 4 bytes */

src/intel/compiler/brw_nir.c

@@ -1420,7 +1420,7 @@ brw_nir_should_vectorize_mem(unsigned align_mul, unsigned align_offset,
     * those back into 32-bit ones anyway and UBO loads aren't split in NIR so
     * we don't want to make a mess for the back-end.
     */
-   if (bit_size > 32 || hole_size)
+   if (bit_size > 32 || hole_size || !nir_num_components_valid(num_components))
       return false;
 
    if (low->intrinsic == nir_intrinsic_load_ubo_uniform_block_intel ||

src/intel/compiler/elk/elk_nir.c

@@ -1137,7 +1137,7 @@ elk_nir_should_vectorize_mem(unsigned align_mul, unsigned align_offset,
     * those back into 32-bit ones anyway and UBO loads aren't split in NIR so
     * we don't want to make a mess for the back-end.
     */
-   if (bit_size > 32 || hole_size)
+   if (bit_size > 32 || hole_size || !nir_num_components_valid(num_components))
       return false;
 
    if (low->intrinsic == nir_intrinsic_load_ubo_uniform_block_intel ||