24.0624.06

5 files changed, 62 insertions, 24 deletions

diff --git a/C/7zVersion.h b/C/7zVersion.h
index 72b915a..75052e9 100644
--- a/C/7zVersion.h
+++ b/C/7zVersion.h
@@ -1,7 +1,7 @@
 #define MY_VER_MAJOR 24
-#define MY_VER_MINOR 05
+#define MY_VER_MINOR 06
 #define MY_VER_BUILD 0
-#define MY_VERSION_NUMBERS "24.05"
+#define MY_VERSION_NUMBERS "24.06"
 #define MY_VERSION MY_VERSION_NUMBERS
 
 #ifdef MY_CPU_NAME
@@ -10,7 +10,7 @@
   #define MY_VERSION_CPU MY_VERSION
 #endif
 
-#define MY_DATE "2024-05-14"
+#define MY_DATE "2024-05-26"
 #undef MY_COPYRIGHT
 #undef MY_VERSION_COPYRIGHT_DATE
 #define MY_AUTHOR_NAME "Igor Pavlov"
diff --git a/C/Blake2s.c b/C/Blake2s.c
index 459e76b..abb907d 100644
--- a/C/Blake2s.c
+++ b/C/Blake2s.c
@@ -1,5 +1,5 @@
 /* Blake2s.c -- BLAKE2sp Hash
-2024-01-29 : Igor Pavlov : Public domain
+2024-05-18 : Igor Pavlov : Public domain
 2015-2019 : Samuel Neves : original code : CC0 1.0 Universal (CC0 1.0). */
 
 #include "Precomp.h"
@@ -12,6 +12,17 @@
 #include "Compiler.h"
 #include "CpuArch.h"
 
+/*
+  if defined(__AVX512F__) && defined(__AVX512VL__)
+  {
+    we define Z7_BLAKE2S_USE_AVX512_ALWAYS,
+    but the compiler can use avx512 for any code.
+  }
+  else if defined(Z7_BLAKE2S_USE_AVX512_ALWAYS)
+    { we use avx512 only for sse* and avx* branches of code. }
+*/
+// #define Z7_BLAKE2S_USE_AVX512_ALWAYS // for debug
+
 #if defined(__SSE2__)
     #define Z7_BLAKE2S_USE_VECTORS
 #elif defined(MY_CPU_X86_OR_AMD64)
@@ -59,6 +70,9 @@
 #endif // SSSE3
 
 #if defined(__GNUC__) || defined(__clang__)
+#if defined(Z7_BLAKE2S_USE_AVX512_ALWAYS) && !(defined(__AVX512F__) && defined(__AVX512VL__))
+    #define BLAKE2S_ATTRIB_128BIT  __attribute__((__target__("avx512vl,avx512f")))
+#else
   #if defined(Z7_BLAKE2S_USE_SSE41)
     #define BLAKE2S_ATTRIB_128BIT  __attribute__((__target__("sse4.1")))
   #elif defined(Z7_BLAKE2S_USE_SSSE3)
@@ -67,6 +81,7 @@
     #define BLAKE2S_ATTRIB_128BIT  __attribute__((__target__("sse2")))
   #endif
 #endif
+#endif
 
 
 #if defined(__AVX2__)
@@ -77,7 +92,11 @@
       || defined(Z7_LLVM_CLANG_VERSION) && (Z7_LLVM_CLANG_VERSION >= 30100)
     #define Z7_BLAKE2S_USE_AVX2
     #ifdef Z7_BLAKE2S_USE_AVX2
+#if defined(Z7_BLAKE2S_USE_AVX512_ALWAYS) && !(defined(__AVX512F__) && defined(__AVX512VL__))
+      #define BLAKE2S_ATTRIB_AVX2  __attribute__((__target__("avx512vl,avx512f")))
+#else
       #define BLAKE2S_ATTRIB_AVX2  __attribute__((__target__("avx2")))
+#endif
     #endif
   #elif  defined(Z7_MSC_VER_ORIGINAL) && (Z7_MSC_VER_ORIGINAL >= 1800) \
       || defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1400)
@@ -107,7 +126,9 @@
 
 #if defined(__AVX512F__) && defined(__AVX512VL__)
    // && defined(Z7_MSC_VER_ORIGINAL) && (Z7_MSC_VER_ORIGINAL > 1930)
+  #ifndef Z7_BLAKE2S_USE_AVX512_ALWAYS
   #define Z7_BLAKE2S_USE_AVX512_ALWAYS
+  #endif
   // #pragma message ("=== Blake2s AVX512")
 #endif
 
@@ -1164,7 +1185,9 @@ Blake2sp_Final_V128_Fast(UInt32 *states)
 #if 1 && defined(Z7_BLAKE2S_USE_AVX512_ALWAYS)
   #define MM256_ROR_EPI32  _mm256_ror_epi32
   #define Z7_MM256_ROR_EPI32_IS_SUPPORTED
+#ifdef Z7_BLAKE2S_USE_AVX2_WAY2
   #define LOAD_ROTATE_CONSTS_256
+#endif
 #else
 #ifdef Z7_BLAKE2S_USE_AVX2_WAY_SLOW
 #ifdef Z7_BLAKE2S_USE_AVX2_WAY2
@@ -2549,9 +2572,11 @@ void z7_Black2sp_Prepare(void)
 
 #if defined(MY_CPU_X86_OR_AMD64)
     #if defined(Z7_BLAKE2S_USE_AVX512_ALWAYS)
+      // optional check
+      #if 0 || !(defined(__AVX512F__) && defined(__AVX512VL__))
       if (CPU_IsSupported_AVX512F_AVX512VL())
-    #endif
-    #if defined(Z7_BLAKE2S_USE_SSE41)
+      #endif
+    #elif defined(Z7_BLAKE2S_USE_SSE41)
       if (CPU_IsSupported_SSE41())
     #elif defined(Z7_BLAKE2S_USE_SSSE3)
       if (CPU_IsSupported_SSSE3())
@@ -2584,12 +2609,14 @@ void z7_Black2sp_Prepare(void)
 
 #ifdef Z7_BLAKE2S_USE_AVX2
 #if defined(MY_CPU_X86_OR_AMD64)
-    if (
-    #if 0 && defined(Z7_BLAKE2S_USE_AVX512_ALWAYS)
-        CPU_IsSupported_AVX512F_AVX512VL() &&
+    
+    #if defined(Z7_BLAKE2S_USE_AVX512_ALWAYS)
+      #if 0
+        if (CPU_IsSupported_AVX512F_AVX512VL())
+      #endif
+    #else
+        if (CPU_IsSupported_AVX2())
     #endif
-        CPU_IsSupported_AVX2()
-        )
 #endif
     {
     // #pragma message ("=== Blake2s AVX2")
diff --git a/C/CpuArch.c b/C/CpuArch.c
index d51b38a..c131a68 100644
--- a/C/CpuArch.c
+++ b/C/CpuArch.c
@@ -1,5 +1,5 @@
 /* CpuArch.c -- CPU specific code
-2024-03-02 : Igor Pavlov : Public domain */
+2024-05-18 : Igor Pavlov : Public domain */
 
 #include "Precomp.h"
 
@@ -638,7 +638,7 @@ BoolInt CPU_IsSupported_AVX(void)
 
   {
     const UInt32 bm = (UInt32)x86_xgetbv_0(MY_XCR_XFEATURE_ENABLED_MASK);
-    // printf("\n=== XGetBV=%d\n", bm);
+    // printf("\n=== XGetBV=0x%x\n", bm);
     return 1
         & (BoolInt)(bm >> 1)  // SSE state is supported (set by OS) for storing/restoring
         & (BoolInt)(bm >> 2); // AVX state is supported (set by OS) for storing/restoring
@@ -662,8 +662,7 @@ BoolInt CPU_IsSupported_AVX2(void)
   }
 }
 
-/*
-// fix it:
+#if 0
 BoolInt CPU_IsSupported_AVX512F_AVX512VL(void)
 {
   if (!CPU_IsSupported_AVX())
@@ -672,14 +671,25 @@ BoolInt CPU_IsSupported_AVX512F_AVX512VL(void)
     return False;
   {
     UInt32 d[4];
+    BoolInt v;
     z7_x86_cpuid(d, 7);
     // printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]);
+    v = 1
+      & (BoolInt)(d[1] >> 16)  // avx512f
+      & (BoolInt)(d[1] >> 31); // avx512vl
+    if (!v)
+      return False;
+  }
+  {
+    const UInt32 bm = (UInt32)x86_xgetbv_0(MY_XCR_XFEATURE_ENABLED_MASK);
+    // printf("\n=== XGetBV=0x%x\n", bm);
     return 1
-      & (BoolInt)(d[1] >> 16)  // avx512-f
-      & (BoolInt)(d[1] >> 31); // avx512-Vl
+        & (BoolInt)(bm >> 5)  // OPMASK
+        & (BoolInt)(bm >> 6)  // ZMM upper 256-bit
+        & (BoolInt)(bm >> 7); // ZMM16 ... ZMM31
   }
 }
-*/
+#endif
 
 BoolInt CPU_IsSupported_VAES_AVX2(void)
 {
diff --git a/C/CpuArch.h b/C/CpuArch.h
index dfc68f1..d632c2b 100644
--- a/C/CpuArch.h
+++ b/C/CpuArch.h
@@ -1,5 +1,5 @@
 /* CpuArch.h -- CPU specific code
-2024-05-13 : Igor Pavlov : Public domain */
+2024-05-18 : Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_CPU_ARCH_H
 #define ZIP7_INC_CPU_ARCH_H
@@ -370,12 +370,12 @@ MY_CPU_64BIT means that processor can work with 64-bit registers.
 #define Z7_CPU_FAST_BSWAP_SUPPORTED
 
 /* GCC can generate slow code that calls function for __builtin_bswap32() for:
-     - GCC for RISCV, if Zbb extension is not used.
+     - GCC for RISCV, if Zbb/XTHeadBb extension is not used.
      - GCC for SPARC.
    The code from CLANG for SPARC also is not fastest.
    So we don't define Z7_CPU_FAST_BSWAP_SUPPORTED in some cases.
 */
-#elif (!defined(MY_CPU_RISCV) || defined (__riscv_zbb)) \
+#elif (!defined(MY_CPU_RISCV) || defined (__riscv_zbb) || defined(__riscv_xtheadbb)) \
     && !defined(MY_CPU_SPARC) \
     && ( \
        (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))) \
@@ -607,7 +607,7 @@ UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void);
 BoolInt CPU_IsSupported_AES(void);
 BoolInt CPU_IsSupported_AVX(void);
 BoolInt CPU_IsSupported_AVX2(void);
-// BoolInt CPU_IsSupported_AVX512F_AVX512VL(void);
+BoolInt CPU_IsSupported_AVX512F_AVX512VL(void);
 BoolInt CPU_IsSupported_VAES_AVX2(void);
 BoolInt CPU_IsSupported_CMOV(void);
 BoolInt CPU_IsSupported_SSE(void);
diff --git a/C/ZstdDec.c b/C/ZstdDec.c
index ecf6d22..ac159d6 100644
--- a/C/ZstdDec.c
+++ b/C/ZstdDec.c
@@ -1,5 +1,5 @@
 /* ZstdDec.c -- Zstd Decoder
-2024-01-21 : the code was developed by Igor Pavlov, using Zstandard format
+2024-05-26 : the code was developed by Igor Pavlov, using Zstandard format
              specification and original zstd decoder code as reference code.
 original zstd decoder code: Copyright (c) Facebook, Inc. All rights reserved.
 This source code is licensed under BSD 3-Clause License.
@@ -2507,6 +2507,7 @@ SRes ZstdDec1_DecodeBlock(CZstdDec1 *p,
     if (vars.numSeqs == 0)
     {
       p->winPos += numLits;
+      UPDATE_TOTAL_OUT(p, numLits)
       return SZ_OK;
     }
   }
@@ -3310,11 +3311,11 @@ static SRes ZstdDec_DecodeBlock(CZstdDec * const p, CZstdDecState * const ds,
         {
           const SizeT xxh64_winPos = p->decoder.winPos - ZstdDec_GET_UNPROCESSED_XXH64_SIZE(p);
           p->decoder.winPos += outCur;
+          UPDATE_TOTAL_OUT(&p->decoder, outCur)
           p->contentProcessed += outCur;
           ZstdDec_Update_XXH(p, xxh64_winPos);
         }
         // ds->winPos = p->decoder.winPos;  // the caller does it instead. for debug:
-        UPDATE_TOTAL_OUT(&p->decoder, outCur)
         ds->outProcessed += outCur;
         if (p->blockSize -= (UInt32)outCur)
         {