JIT compiler update

Author: Zoltan Herczeg

src/pcre2_jit_compile.c

@@ -2272,7 +2272,7 @@ int i;
 for (i = 0; i < RECURSE_TMP_REG_COUNT; i++)
   {
   SLJIT_ASSERT(status->tmp_regs[i] >= 0);
-  SLJIT_ASSERT(sljit_get_register_index(SLJIT_INT_REGISTER, status->saved_tmp_regs[i]) < 0 || status->tmp_regs[i] == status->saved_tmp_regs[i]);
+  SLJIT_ASSERT(sljit_get_register_index(SLJIT_GP_REGISTER, status->saved_tmp_regs[i]) < 0 || status->tmp_regs[i] == status->saved_tmp_regs[i]);
 
   status->store_bases[i] = -1;
   }
@@ -2292,7 +2292,7 @@ SLJIT_ASSERT(load_base > 0 && store_base > 0);
 if (status->store_bases[next_tmp_reg] == -1)
   {
   /* Preserve virtual registers. */
-  if (sljit_get_register_index(SLJIT_INT_REGISTER, status->saved_tmp_regs[next_tmp_reg]) < 0)
+  if (sljit_get_register_index(SLJIT_GP_REGISTER, status->saved_tmp_regs[next_tmp_reg]) < 0)
     OP1(SLJIT_MOV, status->saved_tmp_regs[next_tmp_reg], 0, tmp_reg, 0);
   }
 else
@@ -2321,7 +2321,7 @@ for (i = 0; i < RECURSE_TMP_REG_COUNT; i++)
     OP1(SLJIT_MOV, SLJIT_MEM1(status->store_bases[next_tmp_reg]), status->store_offsets[next_tmp_reg], tmp_reg, 0);
 
     /* Restore virtual registers. */
-    if (sljit_get_register_index(SLJIT_INT_REGISTER, saved_tmp_reg) < 0)
+    if (sljit_get_register_index(SLJIT_GP_REGISTER, saved_tmp_reg) < 0)
       OP1(SLJIT_MOV, tmp_reg, 0, saved_tmp_reg, 0);
     }
 
@@ -3250,7 +3250,7 @@ if (size == sizeof(sljit_sw))
   return;
   }
 
-if (sljit_get_register_index(SLJIT_INT_REGISTER, TMP3) >= 0 && !sljit_has_cpu_feature(SLJIT_HAS_ZERO_REGISTER))
+if (sljit_get_register_index(SLJIT_GP_REGISTER, TMP3) >= 0 && !sljit_has_cpu_feature(SLJIT_HAS_ZERO_REGISTER))
   {
   OP1(SLJIT_MOV, TMP3, 0, SLJIT_IMM, 0);
   src = TMP3;
@@ -13739,9 +13739,9 @@ jump_list *reqcu_not_found = NULL;
 SLJIT_ASSERT(tables);
 
 #if HAS_VIRTUAL_REGISTERS == 1
-SLJIT_ASSERT(sljit_get_register_index(SLJIT_INT_REGISTER, TMP3) < 0 && sljit_get_register_index(SLJIT_INT_REGISTER, ARGUMENTS) < 0 && sljit_get_register_index(SLJIT_INT_REGISTER, RETURN_ADDR) < 0);
+SLJIT_ASSERT(sljit_get_register_index(SLJIT_GP_REGISTER, TMP3) < 0 && sljit_get_register_index(SLJIT_GP_REGISTER, ARGUMENTS) < 0 && sljit_get_register_index(SLJIT_GP_REGISTER, RETURN_ADDR) < 0);
 #elif HAS_VIRTUAL_REGISTERS == 0
-SLJIT_ASSERT(sljit_get_register_index(SLJIT_INT_REGISTER, TMP3) >= 0 && sljit_get_register_index(SLJIT_INT_REGISTER, ARGUMENTS) >= 0 && sljit_get_register_index(SLJIT_INT_REGISTER, RETURN_ADDR) >= 0);
+SLJIT_ASSERT(sljit_get_register_index(SLJIT_GP_REGISTER, TMP3) >= 0 && sljit_get_register_index(SLJIT_GP_REGISTER, ARGUMENTS) >= 0 && sljit_get_register_index(SLJIT_GP_REGISTER, RETURN_ADDR) >= 0);
 #else
 #error "Invalid value for HAS_VIRTUAL_REGISTERS"
 #endif

src/pcre2_jit_simd_inc.h

@@ -201,7 +201,7 @@ switch (step)
   }
 }
 
-#define JIT_HAS_FAST_FORWARD_CHAR_SIMD (sljit_has_cpu_feature(SLJIT_HAS_SSE2))
+#define JIT_HAS_FAST_FORWARD_CHAR_SIMD (sljit_has_cpu_feature(SLJIT_HAS_FPU))
 
 static void fast_forward_char_simd(compiler_common *common, PCRE2_UCHAR char1, PCRE2_UCHAR char2, sljit_s32 offset)
 {
@@ -214,8 +214,8 @@ struct sljit_label *restart;
 struct sljit_jump *quit;
 struct sljit_jump *partial_quit[2];
 vector_compare_type compare_type = vector_compare_match1;
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, STR_PTR);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, STR_PTR);
 sljit_s32 data_ind = 0;
 sljit_s32 tmp_ind = 1;
 sljit_s32 cmp1_ind = 2;
@@ -365,7 +365,7 @@ if (common->utf && offset > 0)
 #endif
 }
 
-#define JIT_HAS_FAST_REQUESTED_CHAR_SIMD (sljit_has_cpu_feature(SLJIT_HAS_SSE2))
+#define JIT_HAS_FAST_REQUESTED_CHAR_SIMD (sljit_has_cpu_feature(SLJIT_HAS_FPU))
 
 static jump_list *fast_requested_char_simd(compiler_common *common, PCRE2_UCHAR char1, PCRE2_UCHAR char2)
 {
@@ -375,8 +375,8 @@ struct sljit_label *start;
 struct sljit_jump *quit;
 jump_list *not_found = NULL;
 vector_compare_type compare_type = vector_compare_match1;
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, STR_PTR);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, STR_PTR);
 sljit_s32 data_ind = 0;
 sljit_s32 tmp_ind = 1;
 sljit_s32 cmp1_ind = 2;
@@ -497,7 +497,7 @@ return not_found;
 
 #ifndef _WIN64
 
-#define JIT_HAS_FAST_FORWARD_CHAR_PAIR_SIMD (sljit_has_cpu_feature(SLJIT_HAS_SSE2))
+#define JIT_HAS_FAST_FORWARD_CHAR_PAIR_SIMD (sljit_has_cpu_feature(SLJIT_HAS_FPU))
 
 static void fast_forward_char_pair_simd(compiler_common *common, sljit_s32 offs1,
   PCRE2_UCHAR char1a, PCRE2_UCHAR char1b, sljit_s32 offs2, PCRE2_UCHAR char2a, PCRE2_UCHAR char2b)
@@ -509,9 +509,9 @@ vector_compare_type compare2_type = vector_compare_match1;
 sljit_u32 bit1 = 0;
 sljit_u32 bit2 = 0;
 sljit_u32 diff = IN_UCHARS(offs1 - offs2);
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 tmp2_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP2);
-sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, STR_PTR);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 tmp2_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP2);
+sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, STR_PTR);
 sljit_s32 data1_ind = 0;
 sljit_s32 data2_ind = 1;
 sljit_s32 tmp1_ind = 2;
@@ -1183,7 +1183,7 @@ if (step == 0)
   OP1(SLJIT_MOV, tmp_general_reg, 0, SLJIT_IMM, chr);
 
   /* VLVG */
-  instruction[0] = (sljit_u16)(0xe700 | (dst_vreg << 4) | sljit_get_register_index(SLJIT_INT_REGISTER, tmp_general_reg));
+  instruction[0] = (sljit_u16)(0xe700 | (dst_vreg << 4) | sljit_get_register_index(SLJIT_GP_REGISTER, tmp_general_reg));
   instruction[1] = 0;
   instruction[2] = (sljit_u16)((VECTOR_ELEMENT_SIZE << 12) | (0x8 << 8) | 0x22);
   sljit_emit_op_custom(compiler, instruction, 6);
@@ -1262,8 +1262,8 @@ struct sljit_label *restart;
 struct sljit_jump *quit;
 struct sljit_jump *partial_quit[2];
 vector_compare_type compare_type = vector_compare_match1;
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, STR_PTR);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, STR_PTR);
 sljit_s32 data_ind = 0;
 sljit_s32 tmp_ind = 1;
 sljit_s32 cmp1_ind = 2;
@@ -1460,8 +1460,8 @@ struct sljit_label *start;
 struct sljit_jump *quit;
 jump_list *not_found = NULL;
 vector_compare_type compare_type = vector_compare_match1;
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 tmp3_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP3);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 tmp3_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP3);
 sljit_s32 data_ind = 0;
 sljit_s32 tmp_ind = 1;
 sljit_s32 cmp1_ind = 2;
@@ -1630,9 +1630,9 @@ vector_compare_type compare2_type = vector_compare_match1;
 sljit_u32 bit1 = 0;
 sljit_u32 bit2 = 0;
 sljit_s32 diff = IN_UCHARS(offs2 - offs1);
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 tmp2_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP2);
-sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, STR_PTR);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 tmp2_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP2);
+sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, STR_PTR);
 sljit_s32 data1_ind = 0;
 sljit_s32 data2_ind = 1;
 sljit_s32 tmp1_ind = 2;
@@ -1949,8 +1949,8 @@ struct sljit_label *restart;
 struct sljit_jump *quit;
 struct sljit_jump *partial_quit[2];
 vector_compare_type compare_type = vector_compare_match1;
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, STR_PTR);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, STR_PTR);
 sljit_s32 data_ind = 0;
 sljit_s32 tmp_ind = 1;
 sljit_s32 cmp1_ind = 2;
@@ -2082,8 +2082,8 @@ struct sljit_label *start;
 struct sljit_jump *quit;
 jump_list *not_found = NULL;
 vector_compare_type compare_type = vector_compare_match1;
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, STR_PTR);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, STR_PTR);
 sljit_s32 data_ind = 0;
 sljit_s32 tmp_ind = 1;
 sljit_s32 cmp1_ind = 2;
@@ -2183,9 +2183,9 @@ vector_compare_type compare2_type = vector_compare_match1;
 sljit_u32 bit1 = 0;
 sljit_u32 bit2 = 0;
 sljit_u32 diff = IN_UCHARS(offs1 - offs2);
-sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP1);
-sljit_s32 tmp2_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, TMP2);
-sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_INT_REGISTER, STR_PTR);
+sljit_s32 tmp1_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP1);
+sljit_s32 tmp2_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, TMP2);
+sljit_s32 str_ptr_reg_ind = sljit_get_register_index(SLJIT_GP_REGISTER, STR_PTR);
 sljit_s32 data1_ind = 0;
 sljit_s32 data2_ind = 1;
 sljit_s32 tmp1_ind = 2;

src/sljit/allocator_src/sljitExecAllocatorApple.c

@@ -33,15 +33,18 @@
    On non-macOS systems, returns MAP_JIT if it is defined.
 */
 #include <TargetConditionals.h>
-#if TARGET_OS_OSX
-#if defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86
+
+#if (defined(TARGET_OS_OSX) && TARGET_OS_OSX) || (TARGET_OS_MAC && !TARGET_OS_IPHONE)
+
+#if defined(SLJIT_CONFIG_X86) && SLJIT_CONFIG_X86
+
 #include <sys/utsname.h>
 #include <stdlib.h>
 
 #define SLJIT_MAP_JIT	(get_map_jit_flag())
 #define SLJIT_UPDATE_WX_FLAGS(from, to, enable_exec)
 
-static SLJIT_INLINE int get_map_jit_flag()
+static SLJIT_INLINE int get_map_jit_flag(void)
 {
 	size_t page_size;
 	void *ptr;
@@ -67,10 +70,8 @@ static SLJIT_INLINE int get_map_jit_flag()
 	}
 	return map_jit_flag;
 }
-#else /* !SLJIT_CONFIG_X86 */
-#if !(defined SLJIT_CONFIG_ARM && SLJIT_CONFIG_ARM)
-#error "Unsupported architecture"
-#endif /* SLJIT_CONFIG_ARM */
+
+#elif defined(SLJIT_CONFIG_ARM) && SLJIT_CONFIG_ARM
 
 #include <AvailabilityMacros.h>
 #include <pthread.h>
@@ -86,9 +87,24 @@ static SLJIT_INLINE void apple_update_wx_flags(sljit_s32 enable_exec)
 #endif /* BigSur */
 	pthread_jit_write_protect_np(enable_exec);
 }
-#endif /* SLJIT_CONFIG_X86 */
+
+#elif defined(SLJIT_CONFIG_PPC) && SLJIT_CONFIG_PPC
+
+#define SLJIT_MAP_JIT	(0)
+#define SLJIT_UPDATE_WX_FLAGS(from, to, enable_exec)
+
+#else
+#error "Unsupported architecture"
+#endif /* SLJIT_CONFIG */
+
 #else /* !TARGET_OS_OSX */
+
+#ifdef MAP_JIT
 #define SLJIT_MAP_JIT	(MAP_JIT)
+#else
+#define SLJIT_MAP_JIT	(0)
+#endif
+
 #endif /* TARGET_OS_OSX */
 
 static SLJIT_INLINE void* alloc_chunk(sljit_uw size)

src/sljit/sljitConfigInternal.h

@@ -85,6 +85,8 @@ extern "C" {
    Other macros:
      SLJIT_FUNC : calling convention attribute for both calling JIT from C and C calling back from JIT
      SLJIT_W(number) : defining 64 bit constants on 64 bit architectures (platform independent helper)
+     SLJIT_F64_SECOND(reg) : provides the register index of the second 32 bit part of a 64 bit
+                             floating point register when SLJIT_HAS_F64_AS_F32_PAIR returns non-zero
 */
 
 /***********************************************************/
@@ -201,6 +203,10 @@ extern "C" {
 /* Instruction cache flush. */
 /****************************/
 
+#ifdef __APPLE__
+#include <AvailabilityMacros.h>
+#endif
+
 /*
  * TODO:
  *
@@ -241,7 +247,7 @@ extern "C" {
 /* Not required to implement on archs with unified caches. */
 #define SLJIT_CACHE_FLUSH(from, to)
 
-#elif defined __APPLE__
+#elif defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED >= 1050
 
 /* Supported by all macs since Mac OS 10.5.
    However, it does not work on non-jailbroken iOS devices,
@@ -393,9 +399,10 @@ typedef double sljit_f64;
 /* Auto detecting mips revision. */
 #if (defined __mips_isa_rev) && (__mips_isa_rev >= 6)
 #define SLJIT_MIPS_REV 6
-#elif (defined __mips_isa_rev && __mips_isa_rev >= 1) \
-	|| (defined __clang__ && defined _MIPS_ARCH_OCTEON) \
-	|| (defined __clang__ && defined _MIPS_ARCH_P5600)
+#elif defined(__mips_isa_rev) && __mips_isa_rev >= 1
+#define SLJIT_MIPS_REV __mips_isa_rev
+#elif defined(__clang__) \
+	&& (defined(_MIPS_ARCH_OCTEON) || defined(_MIPS_ARCH_P5600))
 /* clang either forgets to define (clang-7) __mips_isa_rev at all
  * or sets it to zero (clang-8,-9) for -march=octeon (MIPS64 R2+)
  * and -march=p5600 (MIPS32 R5).
@@ -676,6 +683,19 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_sw sljit_exec_offset(void* ptr);
 #define SLJIT_HAS_STATUS_FLAGS_STATE 1
 #endif
 
+/***************************************/
+/* Floating point register management. */
+/***************************************/
+
+#if (defined SLJIT_CONFIG_ARM_32 && SLJIT_CONFIG_ARM_32) \
+	|| (defined SLJIT_CONFIG_MIPS_32 && SLJIT_CONFIG_MIPS_32)
+#define SLJIT_F64_SECOND(reg) \
+	((reg) + SLJIT_FS0)
+#else /* !SLJIT_CONFIG_ARM_32 && !SLJIT_CONFIG_MIPS_32 */
+#define SLJIT_F64_SECOND(reg) \
+	(reg)
+#endif /* SLJIT_CONFIG_ARM_32 || SLJIT_CONFIG_MIPS_32 */
+
 /*************************************/
 /* Debug and verbose related macros. */
 /*************************************/