Merge pull request #4333 from matrix/fix_apple_intel_rp_apple_metal_parallel_comp

Fixed bug in inc_rp_optimized.cl on Apple Intel with Metal

OpenCL/inc_common.h

@@ -304,6 +304,8 @@ DECLSPEC void hc_enc_init (PRIVATE_AS hc_enc_t *hc_enc);
 DECLSPEC int hc_enc_has_next (PRIVATE_AS hc_enc_t *hc_enc, const int sz);
 DECLSPEC int hc_enc_next (PRIVATE_AS hc_enc_t *hc_enc, PRIVATE_AS const u32 *src_buf, const int src_len, const int src_sz, PRIVATE_AS u32 *dst_buf, const int dst_sz);
 DECLSPEC int hc_enc_next_global (PRIVATE_AS hc_enc_t *hc_enc, GLOBAL_AS const u32 *src_buf, const int src_len, const int src_sz, PRIVATE_AS u32 *dst_buf, const int dst_sz);
+DECLSPEC int hc_enc_validate_utf8 (PRIVATE_AS const u32 *src_buf, const int src_pos, const int extraBytesToRead);
+DECLSPEC int hc_enc_validate_utf8_global (GLOBAL_AS const u32 *src_buf, const int src_pos, const int extraBytesToRead);
 
 DECLSPEC int pkcs_padding_bs8 (PRIVATE_AS const u32 *data_buf, const int data_len);
 DECLSPEC int pkcs_padding_bs16 (PRIVATE_AS const u32 *data_buf, const int data_len);
@@ -359,6 +361,7 @@ DECLSPEC void append_0x01_2x4_S (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, const u
 DECLSPEC void append_0x06_2x4_S (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, const u32 offset);
 DECLSPEC void append_0x01_4x4_S (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, const u32 offset);
 DECLSPEC void append_0x2d_4x4_S (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, const u32 offset);
+DECLSPEC void append_0x3a_4x4_S (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, PRIVATE_AS u32 *w3, const u32 offset);
 DECLSPEC void append_0x80_1x4_S (PRIVATE_AS u32 *w0, const u32 offset);
 DECLSPEC void append_0x80_2x4_S (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, const u32 offset);
 DECLSPEC void append_0x80_3x4_S (PRIVATE_AS u32 *w0, PRIVATE_AS u32 *w1, PRIVATE_AS u32 *w2, const u32 offset);
@@ -386,5 +389,6 @@ DECLSPEC void append_0x06_2x4_VV (PRIVATE_AS u32x *w0, PRIVATE_AS u32x *w1, cons
 DECLSPEC void append_0x80_2x4_VV (PRIVATE_AS u32x *w0, PRIVATE_AS u32x *w1, const u32x offset);
 DECLSPEC void append_0x80_4x4_VV (PRIVATE_AS u32x *w0, PRIVATE_AS u32x *w1, PRIVATE_AS u32x *w2, PRIVATE_AS u32x *w3, const u32x offset);
 DECLSPEC void append_0x2d_4x4_VV (PRIVATE_AS u32x *w0, PRIVATE_AS u32x *w1, PRIVATE_AS u32x *w2, PRIVATE_AS u32x *w3, const u32x offset);
+DECLSPEC void append_0x3a_4x4_VV (PRIVATE_AS u32x *w0, PRIVATE_AS u32x *w1, PRIVATE_AS u32x *w2, PRIVATE_AS u32x *w3, const u32x offset);
 
 #endif // INC_COMMON_H

OpenCL/inc_rp_optimized.cl

@@ -1084,9 +1084,18 @@ DECLSPEC HC_INLINE_RP u32 rule_op_mangle_toggle_at_sep (MAYBE_UNUSED const u32 p
       {
         ro = 1 << i;
 
-        break;
+        #if defined(IS_METAL) && !defined(IS_APPLE_SILICON)
-      }
 
+        i = 32;
+
+        continue;
+
+        #else
+
+        break; // bug on Apple Intel with Metal
+
+        #endif
+      }
       occurence++;
     }
   }
@@ -2235,6 +2244,8 @@ DECLSPEC u32 apply_rule_optimized (const u32 name, const u32 p0, const u32 p1, P
 {
   u32 out_len = in_len;
 
+  if (name == RULE_OP_MANGLE_NOOP) return out_len;
+
   switch (name)
   {
     case RULE_OP_MANGLE_LREST:            out_len = rule_op_mangle_lrest            (p0, p1, buf0, buf1, out_len); break;
@@ -2284,6 +2295,7 @@ DECLSPEC u32 apply_rule_optimized (const u32 name, const u32 p0, const u32 p1, P
   return out_len;
 }
 
+//DECLSPEC u32 apply_rules_optimized (PRIVATE_AS const u32 *cmds, PRIVATE_AS u32 *buf0, PRIVATE_AS u32 *buf1, const u32 len)
 DECLSPEC u32 apply_rules_optimized (CONSTANT_AS const u32 *cmds, PRIVATE_AS u32 *buf0, PRIVATE_AS u32 *buf1, const u32 len)
 {
   u32 out_len = len;

OpenCL/inc_rp_optimized.h

@@ -14,7 +14,7 @@
 #define MAYBE_UNUSED
 #endif
 
-#ifdef IS_APPLE_SILICON
+#if defined(IS_METAL) && !defined(IS_APPLE_SILICON)
 #define HC_INLINE_RP __attribute__ ((noinline))
 #else
 #define HC_INLINE_RP

OpenCL/inc_shared.h

@@ -63,4 +63,6 @@
 
 #endif // IS_METAL
 
+DECLSPEC void gpu_decompress_entry (GLOBAL_AS pw_idx_t *pws_idx, GLOBAL_AS u32 *pws_comp, PRIVATE_AS pw_t *buf, const u64 gid);
+
 #endif // INC_SHARED_H

OpenCL/m00000_a0-optimized.cl

@@ -18,12 +18,6 @@
 
 KERNEL_FQ KERNEL_FA void m00000_m04 (KERN_ATTR_RULES ())
 {
-  /**
-   * modifier
-   */
-
-  const u64 lid = get_local_id (0);
-
   /**
    * base
    */
@@ -153,12 +147,6 @@ KERNEL_FQ KERNEL_FA void m00000_m16 (KERN_ATTR_RULES ())
 
 KERNEL_FQ KERNEL_FA void m00000_s04 (KERN_ATTR_RULES ())
 {
-  /**
-   * modifier
-   */
-
-  const u64 lid = get_local_id (0);
-
   /**
    * base
    */

docs/changes.txt

@@ -91,6 +91,7 @@
 - Added verification of token buffer length when using TOKEN_ATTR_FIXED_LENGTH
 - Fixed a bug in all SCRYPT-based hash modes with Apple Metal
 - Fixed buffer overflow on module_26600.c / module_hash_encode()
+- Fixed bug in inc_rp_optimized.cl on Apple Intel with Metal
 - Fixed bug in module_constraints and kernel for hash-mode 7801
 - Fixed bug in module_constraints and kernel for hash-mode 7800
 - Fixed bug in 18400 module_hash_encode
@@ -175,9 +176,10 @@
 - Metal Backend: added workaround to set the true Processor value in Metal devices on Apple Intel
 - Metal Backend: added support to 2D/3D Compute
 - Metal Backend: allow use of devices with Metal if runtime version is >= 200
+- Metal Backend: define USE_BITSELECT and USE_SWIZZLE for Apple Intel
 - Metal Backend: disable Metal devices only if at least one OpenCL device is active
 - Metal Backend: improved compute workloads calculation
-- Metal Backend: define USE_BITSELECT and USE_SWIZZLE for Apple Intel
+- Metal Backend: parallelize pipeline state object (PSO) compilation internally
 - Modules: Check UnpackSize to raise false positive with hc_decompress_rar
 - User Options: added --metal-compiler-runtime option
 - User Options: limit --bitmap-max value to 31

src/ext_metal.m

@@ -11,6 +11,7 @@
 #include "ext_metal.h"
 
 #include <sys/sysctl.h>
+#include <objc/message.h>
 
 #include <CoreFoundation/CoreFoundation.h>
 #include <Foundation/Foundation.h>
@@ -309,6 +310,13 @@ int hc_mtlDeviceGet (void *hashcat_ctx, mtl_device_id *metal_device, int ordinal
     return -1;
   }
 
+  // parallelize pipeline state object (PSO) compilation internally
+
+  if ([device respondsToSelector:@selector(setShouldMaximizeConcurrentCompilation:)])
+  {
+    ((void (*)(id, SEL, BOOL))objc_msgSend)(device, @selector(setShouldMaximizeConcurrentCompilation:), YES);
+  }
+
   *metal_device = device;
 
   return 0;
@@ -1440,6 +1448,7 @@ int hc_mtlCreateLibraryWithSource (void *hashcat_ctx, mtl_device_id metal_device
       }
 
       compileOptions.preprocessorMacros = build_options_dict;
+
       /*
       compileOptions.optimizationLevel = MTLLibraryOptimizationLevelSize;
       compileOptions.mathMode = MTLMathModeSafe;