diff --git a/AUTHORS b/AUTHORS
index 6bd323d2c..1fb91adaf 100644
--- a/AUTHORS
+++ b/AUTHORS
@@ -243,6 +243,7 @@ Thanar2
 thaspel
 theo77186
 TierynnB
+Timothy Herchen (anematode)
 Ting-Hsuan Huang (fffelix-huang)
 Tobias Steinmann
 Tomasz Sobczyk (Sopel97)
diff --git a/src/nnue/layers/affine_transform_sparse_input.h b/src/nnue/layers/affine_transform_sparse_input.h
index a073c6196..11e460666 100644
--- a/src/nnue/layers/affine_transform_sparse_input.h
+++ b/src/nnue/layers/affine_transform_sparse_input.h
@@ -22,6 +22,7 @@
 #define NNUE_LAYERS_AFFINE_TRANSFORM_SPARSE_INPUT_H_INCLUDED
 
 #include <algorithm>
+#include <cstddef>
 #include <cstdint>
 #include <iostream>
 
@@ -287,12 +288,18 @@ class AffineTransformSparseInput {
         for (IndexType k = 0; k < NumRegs; ++k)
             acc[k] = biasvec[k];
 
-        for (IndexType j = 0; j < count; ++j)
+        auto* start = nnz;
+        auto* end   = nnz + count;
+
+        // convince GCC to not do weird pointer arithmetic in the following loop
+        const std::int8_t* weights_cp = weights;
+
+        while (start < end)
         {
-            const auto    i  = nnz[j];
-            const invec_t in = vec_set_32(input32[i]);
-            const auto    col =
-              reinterpret_cast<const invec_t*>(&weights[i * OutputDimensions * ChunkSize]);
+            const std::ptrdiff_t i = *start;
+            start++;
+            const invec_t in  = vec_set_32(input32[i]);
+            const auto    col = (const invec_t*) (&weights_cp[i * OutputDimensions * ChunkSize]);
             for (IndexType k = 0; k < NumRegs; ++k)
                 vec_add_dpbusd_32(acc[k], in, col[k]);
         }