PERF: Enlarge zlib read buffer in NIfTI reader for NAS performance (#1634)

The NIfTI reader opened files with a bare gzopen(), leaving zlib at its
8 KB default internal buffer. zlib refills from the underlying file in
buffer-sized chunks, so a multi-GB .nii.gz triggered hundreds of
thousands of small reads. On network attached storage each refill is a
latency-bound round-trip, making reads disproportionately slow compared
to a local disk.

Author: imikejackson

src/Plugins/SimplnxCore/src/SimplnxCore/Filters/Algorithms/ReadNIfTIFile.cpp

@@ -378,6 +378,12 @@ Result<> ReadNIfTIFile::operator()()
     return MakeErrorResult(-34720, fmt::format("Could not open NIfTI file for reading: '{}'", m_InputValues->InputFilePath.string()));
   }
 
+  // Enlarge zlib's internal buffer before the first read/seek so the file is
+  // refilled in large chunks instead of zlib's 8 KB default. This drastically
+  // reduces the number of latency-bound network round-trips when the input
+  // lives on a NAS. See nifti::k_GzReadBufferSize for the rationale.
+  gzbuffer(gz, static_cast<unsigned int>(nx::core::nifti::k_GzReadBufferSize));
+
   const z_off_t targetOffset = static_cast<z_off_t>(md.voxOffset);
   if(gzseek(gz, targetOffset, SEEK_SET) != targetOffset)
   {

src/Plugins/SimplnxCore/src/SimplnxCore/utils/NiftiUtilities.cpp

@@ -119,8 +119,10 @@ Result<NiftiMetadata> ReadNiftiHeader(const std::filesystem::path& filePath, boo
     return MakeErrorResult<NiftiMetadata>(-34700, fmt::format("Could not open NIfTI file for reading: '{}'", pathStr));
   }
 
+  gzbuffer(gz, k_GzReadBufferSize);
+
   nifti_1_header hdr{};
-  int bytesRead = gzread(gz, &hdr, static_cast<unsigned int>(k_HeaderSize));
+  int bytesRead = gzread(gz, &hdr, k_HeaderSize);
   gzclose(gz);
 
   if(bytesRead != static_cast<int>(k_HeaderSize))

src/Plugins/SimplnxCore/src/SimplnxCore/utils/NiftiUtilities.hpp

@@ -26,7 +26,7 @@
  * `byteSwapRequired` from the metadata to do its own stream reads.
  *
  * The helpers here are deliberately independent of the simplnx filter
- * framework so they can be reused from unit tests, a future writer,
+ * framework, so they can be reused from unit tests, a future writer,
  * or a command-line tool.
  */
 
@@ -36,7 +36,7 @@ namespace nx::core::nifti
  * @brief Size of the NIfTI-1 header in bytes. Always 348 for a valid
  *        NIfTI-1 file regardless of byte order.
  */
-inline constexpr usize k_HeaderSize = 348;
+inline constexpr unsigned int k_HeaderSize = 348;
 
 /**
  * @brief Minimum legal value of the `vox_offset` field in a single-file
@@ -46,6 +46,23 @@ inline constexpr usize k_HeaderSize = 348;
  */
 inline constexpr usize k_MinVoxOffset = 352;
 
+/**
+ * @brief Size (in bytes) of zlib's internal input/output buffer, set via
+ *        `gzbuffer()` immediately after every `gzopen()`.
+ *
+ * zlib defaults to an 8 KB internal buffer, which means it refills from
+ * the underlying file in 8 KB chunks of compressed data — one `read()`
+ * syscall each. On a local disk that is negligible, but on network
+ * attached storage (NAS) every refill is a separate network round-trip,
+ * so a multi-GB `.nii.gz` triggers hundreds of thousands of small,
+ * latency-bound reads. Raising the buffer to 4 MiB cuts the number of
+ * round-trips by ~512x while costing only a few × this size in working
+ * memory (negligible next to the voxel volume). Larger values yield
+ * diminishing returns once the buffer exceeds the link's
+ * bandwidth-delay product and the client's own read-ahead.
+ */
+inline constexpr unsigned int k_GzReadBufferSize = 4194304;
+
 /**
  * @brief Magic bytes that identify the single-file `.nii` / `.nii.gz`
  *        format. Trailing null is included per the spec.