improve documentation and update changelog

Author: ixfd64

Changelog-mfakto.txt

@@ -14,7 +14,12 @@ new features:
   - additional code contributions and bug fixes by Danny Chia
 
 enhancements:
-- added RDNA 3 support
+- added support for RDNA 3 devices
+  - also covers RDNA 3.5 and 4
+- improved detection of GPU type
+  - added provisional support for CDNA
+  - implemented pattern-matching code by Rich Felker
+  - additional code contributions by Mingye Wang
 - added code to handle situations where the device's clock speed could not be
   determined
 - improved some warning messages

src/mfakto.ini

@@ -24,22 +24,26 @@
 # GPUType=AUTO      tries to automatically detect the device
 # GPUType=VLIW4     "Northern Islands" GPUs, such as the Radeon HD 6900 series
 # GPUType=VLIW5     "Evergreen" GPUs, such as the Radeon HD 5000 series. Can
-#                   also improve performance on low-end APUs
-# GPUType=GCN       all GPUs based on the GCN architecture. Assumed for unknown
-#                   devices. 1:16 DP
-# GPUType=GCN2      2nd generation GCN architecture (actually, any GCN2/3 with 1:4 or better DP)
-# GPUType=GCN3      3rd generation GCN architecture (actually, any GCN2/3 with improved int32 and 1:8 or better DP)
-# GPUType=GCN4      Polaris GPUs, such as the Radeon RX 460 (actually, GCN 3/4 with bad DP)
-# GPUType=GCN5      14 nm Vega GPUs, such as the Radeon RX Vega 56 (actually, any GCN5 with bad DP)
-# GPUType=GCNF      7 nm Vega GPUs, namely the Vega 20 series (actually, any GCN5 with 1:4 or better DP)
-#                   CDNA is provisionally included                    
-# GPUType=RDNA      devices using the RDNA 1 and 2 microarchitectures, such as
-#                   the Radeon RX 5000 and 6000 series
-# GPUType=RDNA3     devices using the RDNA 3/4 microarchitecture
-# GPUType=APU       VLIW APUs. For old low-end devices, using GPUType=VLIW5 may
-#                   result in better performance.
-# GPUType=CPU       all CPUs. Used when no GPUs are available; also used when
-#                   the '-d c' option is specified
+#                   also improve performance on older APUs
+# GPUType=GCN       the GCN architecture as a whole. Used for unknown devices
+#                   and optimized for 1:16 DP
+# GPUType=GCN2      2nd generation GCN architecture. Can be used for any GCN 2
+#                   or 3 device with 1:4 or higher DP
+# GPUType=GCN3      3rd generation GCN architecture. Can be used for any GCN 2
+#                   or 3 device with fast 32-bit integer multiplication and 1:8
+#                   or higher DP
+# GPUType=GCN4      "Polaris" GPUs, such as the Radeon 400 series. Can be used
+#                   for any GCN 3 or 4 device with low DP
+# GPUType=GCN5      GCN 5 devices with low DP, such as the Vega 10 series
+# GPUType=GCNF      GCN 5 devices with 1:4 or higher DP, such as the Vega 20
+#                   series. Provisionally supports CDNA
+# GPUType=RDNA      RDNA 1 and 2 devices, such as the Radeon RX 5000 and 6000
+#                   series
+# GPUType=RDNA3     RDNA 3 or 4 devices
+# GPUType=APU       APUs with a VLIW-based processor. Older devices may benefit
+#                   more with GPUType=VLIW5
+# GPUType=CPU       all supported CPUs. Used when no GPUs are available, and
+#                   when the '-d c' option is specified
 # GPUType=NVIDIA    most Nvidia devices
 # GPUType=INTEL     Intel Graphics Technology
 #

src/myfnmatch.c

@@ -3,13 +3,13 @@
 /*
  * Lifted from musl commit efa9d396f9d3af6c6f85ec86302b48206c574a38,
  * root/src/regex/fnmatch.c, under a GPLv3-compatible license.
- * 
+ *
  * Mingye Wang 2025: delete a bunch of flags
- * 
+ *
  * musl (www.musl-libc.org)
  * Copyright (C) 2012-2013 Rich Felker
  * SPDX-License-Identifier: MIT
- * 
+ *
  * An implementation of what I call the "Sea of Stars" algorithm for
  * POSIX fnmatch(). The basic idea is that we factor the pattern into
  * a head component (which we match first and can reject without ever
@@ -305,5 +305,4 @@ int fnmatch(const char *pat, const char *str, int flags)
 	return fnmatch_internal(pat, -1, str, -1, flags);
 }
 
-
-#endif
+#endif

src/myfnmatch.h

@@ -7,8 +7,14 @@
 #ifdef __cplusplus
 extern "C" {
 #endif
-/* Define our own. It would be easier to just use ntdll RtlIsNameInExpression() on Windows, but ntdll.lib is Windows 10+ SDK only.
- * There are many license-compatible fnmatch implementations available anyways, so. */
+
+/*
+Define our own. It would be easier to just use NTDLL's RtlIsNameInExpression()
+on Windows, but ntdll.lib is only available in the Native API for Windows 10
+and above.
+
+However, there are many GPL-compatible fnmatch implementations available.
+*/
 #define FNM_CASEFOLD   1
 #define FNM_IGNORECASE FNM_CASEFOLD
 #define FNM_NOMATCH    -1
@@ -26,4 +32,4 @@ inline static int patmatch(const char *pattern, const char *string, int flags)
 #endif
 #endif
 
-#endif
+#endif