Reorganize Chorba activation.

Now WITHOUT_CHORBA will only disable the crc32_chorba C fallback.

SSE2, SSE41 and pclmul variants will still be able to use their Chorba-algorithm based code,
but their fallback to the generic crc32_chorba C code in SSE2 and SSE41 will be disabled,
reducing their performance on really big input buffers (not used during deflate/inflate,
only when calling crc32 directly).

Remove the crc32_c function (and its file crc32_c.c), instead use the normal functable
routing to select between crc32_braid and crc32_chorba.

Disable sse2 and sse4.1 variants of Chorba-crc32 on MSVC older than 2022 due to code
generation bug in 2019 causing segfaults.

Compile either crc32_chorba_small_nondestructive or crc32_chorba_small_nondestructive_32bit,
not both. Don't compile crc32_chorba_32768_nondestructive on 32bit arch.

2 files changed, 17 insertions, 22 deletions

diff --git a/test/benchmarks/benchmark_crc32.cc b/test/benchmarks/benchmark_crc32.cc
index 1e95b27770..3b00f87d72 100644
--- a/test/benchmarks/benchmark_crc32.cc
+++ b/test/benchmarks/benchmark_crc32.cc
@@ -56,12 +56,6 @@ public:
     } \
     BENCHMARK_REGISTER_F(crc32, name)->Arg(1)->Arg(8)->Arg(12)->Arg(16)->Arg(32)->Arg(64)->Arg(512)->Arg(4<<10)->Arg(32<<10)->Arg(256<<10)->Arg(4096<<10);
 
-#ifndef WITHOUT_CHORBA
-BENCHMARK_CRC32(generic_chorba, crc32_c, 1);
-#else
-BENCHMARK_CRC32(generic, crc32_c, 1);
-#endif
-
 BENCHMARK_CRC32(braid, crc32_braid, 1);
 
 #ifdef DISABLE_RUNTIME_CPU_DETECTION
@@ -69,14 +63,16 @@ BENCHMARK_CRC32(native, native_crc32, 1);
 #else
 
 #ifndef WITHOUT_CHORBA
-#   if defined(X86_SSE2) && !defined(NO_CHORBA_SSE)
+BENCHMARK_CRC32(chorba_c, crc32_chorba, 1);
+#endif
+#ifndef WITHOUT_CHORBA_SSE
+#   ifdef X86_SSE2
     BENCHMARK_CRC32(chorba_sse2, crc32_chorba_sse2, test_cpu_features.x86.has_sse2);
-#       if defined(X86_SSE41) && !defined(NO_CHORBA_SSE)
-        BENCHMARK_CRC32(chorba_sse41, crc32_chorba_sse41, test_cpu_features.x86.has_sse41);
-#       endif
+#   endif
+#   ifdef X86_SSE41
+    BENCHMARK_CRC32(chorba_sse41, crc32_chorba_sse41, test_cpu_features.x86.has_sse41);
 #   endif
 #endif
-
 #ifdef ARM_CRC32
 BENCHMARK_CRC32(armv8, crc32_armv8, test_cpu_features.arm.has_crc32);
 #endif
diff --git a/test/test_crc32.cc b/test/test_crc32.cc
index d44d079e97..ca0767d468 100644
--- a/test/test_crc32.cc
+++ b/test/test_crc32.cc
@@ -269,12 +269,6 @@ INSTANTIATE_TEST_SUITE_P(crc32, crc32_variant, testing::ValuesIn(tests));
         hash(func); \
     }
 
-#ifndef WITHOUT_CHORBA
-TEST_CRC32(generic_chorba, crc32_c, 1)
-#else
-TEST_CRC32(generic, crc32_c, 1)
-#endif
-
 TEST_CRC32(braid, crc32_braid, 1)
 
 #ifdef DISABLE_RUNTIME_CPU_DETECTION
@@ -297,6 +291,9 @@ static const int align_offsets[] = {
     }
 #endif
 
+#ifndef WITHOUT_CHORBA
+TEST_CRC32(chorba_c, crc32_chorba, 1)
+#endif
 #ifdef ARM_CRC32
 INSTANTIATE_TEST_SUITE_P(crc32_alignment, crc32_align, testing::ValuesIn(align_offsets));
 TEST_CRC32(armv8, crc32_armv8, test_cpu_features.arm.has_crc32)
@@ -317,11 +314,13 @@ TEST_CRC32(pclmulqdq, crc32_pclmulqdq, test_cpu_features.x86.has_pclmulqdq)
 #ifdef X86_VPCLMULQDQ_CRC
 TEST_CRC32(vpclmulqdq, crc32_vpclmulqdq, (test_cpu_features.x86.has_pclmulqdq && test_cpu_features.x86.has_avx512_common && test_cpu_features.x86.has_vpclmulqdq))
 #endif
-#if !defined(WITHOUT_CHORBA) && defined(X86_SSE2) && !defined(NO_CHORBA_SSE)
-TEST_CRC32(chorba_sse2, crc32_chorba_sse2, test_cpu_features.x86.has_sse2)
-#endif
-#if !defined(WITHOUT_CHORBA) && defined(X86_SSE41) && !defined(NO_CHORBA_SSE)
-TEST_CRC32(chorba_sse41, crc32_chorba_sse41, test_cpu_features.x86.has_sse41)
+#ifndef WITHOUT_CHORBA_SSE
+#   ifdef X86_SSE2
+    TEST_CRC32(chorba_sse2, crc32_chorba_sse2, test_cpu_features.x86.has_sse2)
+#   endif
+#   ifdef X86_SSE41
+    TEST_CRC32(chorba_sse41, crc32_chorba_sse41, test_cpu_features.x86.has_sse41)
+#   endif
 #endif
 #if defined(LOONGARCH_CRC)
 INSTANTIATE_TEST_SUITE_P(crc32_alignment, crc32_align, testing::ValuesIn(align_offsets));