[zlib] Make insert string a little less #ifdef-ie

Remove one level of #ifdef indent to make that part of the code easier
to read. Change the accelerated routine name to end in _simd as is our
way elsewhere in chromium zlib.

Minor: adjust the comments around the performance claims, and move the
important comments re CHROMIUM_ZLIB_NO_CASTAGNOLI into that block.

Bug: 1032721
Change-Id: Icb4044f3b87277d67f0ff004ac70813af0a91f5b
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2148893
Reviewed-by: Chris Blume <cblume@chromium.org>
Reviewed-by: Adenilson Cavalcanti <cavalcantii@chromium.org>
Commit-Queue: Noel Gordon <noel@chromium.org>
Cr-Original-Commit-Position: refs/heads/master@{#760408}
Cr-Mirrored-From: https://chromium.googlesource.com/chromium/src
Cr-Mirrored-Commit: 4520abf183652950cb4b68701f7bd695d127d1f8

1 files changed, 30 insertions, 29 deletions

diff --git a/contrib/optimizations/insert_string.h b/contrib/optimizations/insert_string.h
index 1826601..d3bc33c 100644
--- a/contrib/optimizations/insert_string.h
+++ b/contrib/optimizations/insert_string.h
@@ -4,45 +4,47 @@
  * Use of this source code is governed by a BSD-style license that can be
  * found in the Chromium source repository LICENSE file.
  */
-#ifdef _MSC_VER
+
+#if defined(_MSC_VER)
 #define INLINE __inline
 #else
 #define INLINE inline
 #endif
 
 #include "cpu_features.h"
-/* Optimized insert_string block */
-#if defined(CRC32_SIMD_SSE42_PCLMUL) || defined(CRC32_ARMV8_CRC32)
-#define TARGET_CPU_WITH_CRC
+
 // clang-format off
 #if defined(CRC32_SIMD_SSE42_PCLMUL)
-  /* Required to make MSVC bot build pass. */
-  #include <smmintrin.h>
-  #if defined(__GNUC__) || defined(__clang__)
-    #undef TARGET_CPU_WITH_CRC
+  #include <smmintrin.h>  /* Required to make MSVC bot build pass. */
+
+  #if defined(__clang__) || defined(__GNUC__)
     #define TARGET_CPU_WITH_CRC __attribute__((target("sse4.2")))
+  #else
+    #define TARGET_CPU_WITH_CRC
   #endif
 
   #define _cpu_crc32_u32 _mm_crc32_u32
 
 #elif defined(CRC32_ARMV8_CRC32)
   #if defined(__clang__)
-    #undef TARGET_CPU_WITH_CRC
     #define __crc32cw __builtin_arm_crc32cw
   #endif
 
-  #define _cpu_crc32_u32 __crc32cw
-
   #if defined(__aarch64__)
     #define TARGET_CPU_WITH_CRC __attribute__((target("crc")))
   #else  // !defined(__aarch64__)
     #define TARGET_CPU_WITH_CRC __attribute__((target("armv8-a,crc")))
   #endif  // defined(__aarch64__)
+
+  #define _cpu_crc32_u32 __crc32cw
+
 #endif
 // clang-format on
+
+#if defined(TARGET_CPU_WITH_CRC)
+
 TARGET_CPU_WITH_CRC
-local INLINE Pos insert_string_optimized(deflate_state* const s,
-                                         const Pos str) {
+local INLINE Pos insert_string_simd(deflate_state* const s, const Pos str) {
   Pos ret;
   unsigned *ip, val, h = 0;
 
@@ -64,7 +66,8 @@ local INLINE Pos insert_string_optimized(deflate_state* const s,
   s->prev[str & s->w_mask] = ret;
   return ret;
 }
-#endif /* Optimized insert_string block */
+
+#endif // TARGET_CPU_WITH_CRC
 
 /* ===========================================================================
  * Update a hash value with the given input byte
@@ -99,24 +102,22 @@ local INLINE Pos insert_string_c(deflate_state* const s, const Pos str) {
 }
 
 local INLINE Pos insert_string(deflate_state* const s, const Pos str) {
-/* String dictionary insertion: faster symbol hashing has a positive impact
- * on data compression speeds (around 20% on Intel and 36% on Arm Cortex big
- * cores).
- * A misfeature is that the generated compressed output will differ from
- * vanilla zlib (even though it is still valid 'DEFLATE-d' content).
+/* insert_string_simd string dictionary insertion: this SIMD symbol hashing
+ * significantly improves data compression speed.
  *
- * We offer here a way to disable the optimization if there is the expectation
- * that compressed content should match when compared to vanilla zlib.
+ * Note: the generated compressed output is a valid DEFLATE stream but will
+ * differ from vanilla zlib output ...
  */
-#if !defined(CHROMIUM_ZLIB_NO_CASTAGNOLI)
-  /* TODO(cavalcantii): unify CPU features code. */
-#if defined(CRC32_ARMV8_CRC32)
-  if (arm_cpu_enable_crc32)
-    return insert_string_optimized(s, str);
-#elif defined(CRC32_SIMD_SSE42_PCLMUL)
+#if defined(CHROMIUM_ZLIB_NO_CASTAGNOLI)
+/* ... so this build-time option can used to disable the SIMD symbol hasher
+ * if matching vanilla zlib DEFLATE output is required.
+ */ (;) /* FALLTHOUGH */
+#elif defined(TARGET_CPU_WITH_CRC) && defined(CRC32_SIMD_SSE42_PCLMUL)
   if (x86_cpu_enable_simd)
-    return insert_string_optimized(s, str);
-#endif
+    return insert_string_simd(s, str);
+#elif defined(TARGET_CPU_WITH_CRC) && defined(CRC32_ARMV8_CRC32)
+  if (arm_cpu_enable_crc32)
+    return insert_string_simd(s, str);
 #endif
   return insert_string_c(s, str);
 }