Optimizations

GNUmakefile.os4

@@ -105,7 +105,7 @@ STATIC   := $(if $(STATIC),$(STATIC),yes)
 LARGEDATA :=
 OPTIONS  += $(LARGEDATA) -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -Wa,-mregnames -fno-builtin -nostdlib -D_GNU_SOURCE -D_XOPEN_SOURCE -D_USE_GNU -pipe \
 		-nostdinc -nostartfiles -nostdlib
-OPTIMIZE := -O3 -mregnames -mmultiple -mupdate -ffp-contract=fast -mstrict-align
+OPTIMIZE := -O3 -mregnames -mstrict-align
 
 STABS :=
 DLIBS :=
@@ -117,9 +117,9 @@ else
     DLIBS += $(BUILD_DIR)/lib/libdebug.a
 endif
 
-CFLAGS     := $(WARNINGS) $(OPTIMIZE) $(OPTIONS) $(INCLUDES) -D__USE_INLINE__
-CFLAGS_N   := $(WARNINGS) $(OPTIMIZE) $(OPTIONS) $(INCLUDES)
-AFLAGS     := -Wa,-mregnames -mstrict-align
+CFLAGS     := $(WARNINGS) $(OPTIMIZE) $(OPTIONS) $(INCLUDES) -D__USE_INLINE__ -falign-functions=8
+CFLAGS_N   := $(WARNINGS) $(OPTIMIZE) $(OPTIONS) $(INCLUDES)  -falign-functions=8
+AFLAGS     := -Wa,-mregnames -falign-functions=8 -mstrict-align
 
 ifdef SPE
     CC := ppc-amigaos-gcc-6.4.0

libc.gmk

@@ -576,6 +576,9 @@ C_STRING := \
 	cpu/altivec/vec_memset.o \
 	cpu/altivec/vec_strcpy.o \
 	cpu/generic/bcopy.o \
+	cpu/generic/strcat.o \
+	cpu/generic/strcpy.o \
+	cpu/generic/strlen.o \
 	string/rindex.o \
 	string/stccpy.o \
 	string/stpcpy.o \

library/byteswap/bswap16.c

@@ -1,6 +1,6 @@
 asm("\
 	.text\n\
-	.align 2\n\
+	.align 8\n\
 	.globl bswap16\n\
 	.type	bswap16, @function\n\
 bswap16:\n\

library/byteswap/bswap24.c

@@ -1,5 +1,5 @@
 asm("	.text\n\
-	.align 2\n\
+	.align 8\n\
 	.globl bswap24\n\
 	.type	bswap24, @function\n\
 bswap24:\n\

library/byteswap/bswap32.c

@@ -1,5 +1,5 @@
 asm("	.text\n\
-	.align 2\n\
+	.align 8\n\
 	.globl bswap32\n\
 	.type	bswap32, @function\n\
 bswap32:\n\

library/byteswap/bswap64.c

@@ -1,6 +1,6 @@
 
 asm("	.text\n\
-	.align 2\n\
+	.align 8\n\
 	.globl bswap64\n\
 	.type	bswap64, @function\n\
 bswap64:\n\

library/byteswap/swab.c

@@ -1,6 +1,6 @@
 asm("\
 	.text\n\
-	.align 2\n\
+	.align 8\n\
 	.globl swab\n\
 	.type swab,@function\n\
 swab:\n\

library/byteswap/swab24.c

@@ -1,6 +1,6 @@
 asm("\
 	.text\n\
-	.align 2\n\
+	.align 8\n\
 	.globl swab24\n\
 	.type swab24,@function\n\
 swab24:\n\

library/byteswap/swab32.c

@@ -1,6 +1,6 @@
 asm("\
 	.text\n\
-	.align 2\n\
+	.align 8\n\
 	.globl swab32\n\
 	.type swab32,@function\n\
 swab32:\n\

library/byteswap/swab64.c

@@ -7,7 +7,7 @@
 
 asm("\
 	.text\n\
-	.align 2\n\
+	.align 8\n\
 	.globl swab64\n\
 	.type swab64,@function\n\
 swab64:\n\

library/c.lib_rev.h

@@ -2,7 +2,7 @@
 #define REVISION		0
 #define SUBREVISION		0
 
-#define DATE			"21.03.2024"
+#define DATE			"23.03.2024"
 #define VERS			"clib4.library 1.0.0"
-#define VSTRING			"clib4.library 1.0.0 (21.03.2024)\r\n"
-#define VERSTAG			"\0$VER: clib4.library 1.0.0 (21.03.2024)"
+#define VSTRING			"clib4.library 1.0.0 (23.03.2024)\r\n"
+#define VERSTAG			"\0$VER: clib4.library 1.0.0 (23.03.2024)"

library/cpu/generic/bcopy.S

@@ -60,7 +60,7 @@
 #define	w6	r11
 #define	w7	r0
 #define	w8	r2
-#define	kLong	33					// too long for string ops
+#define	kLong	33				// too long for string ops
 
     .text
 
@@ -69,7 +69,7 @@
     .align 	5
     .global bcopy_g3
 
-bcopy_g3:							// void bcopy(const void *src, void *dst, size_t len)
+bcopy_g3:						// void bcopy(const void *src, void *dst, size_t len)
     cmplwi	rc,kLong			// length > 32 bytes?
     sub		w1,r4,r3			// must move in reverse if (rd-rs)<rc
     mr		rd,r4				// start to move source & dest to canonic spot
@@ -82,8 +82,8 @@ bcopy_g3:							// void bcopy(const void *src, void *dst, size_t len)
     // NB: memcpy() and memmove() must follow bcopy() by 32 bytes, for comm page.
 
     .align	5
-Lmemcpy_g3:							// void* memcpy(void *dst, void *src, size_t len)
-Lmemmove_g3:						// void* memmove(void *dst, const void *src, size_t len)
+Lmemcpy_g3:						// void* memcpy(void *dst, void *src, size_t len)
+Lmemmove_g3:					// void* memmove(void *dst, const void *src, size_t len)
     cmplwi	rc,kLong			// length > 32 bytes?
     sub		w1,r3,rs			// must move in reverse if (rd-rs)<rc
     mr		rd,r3				// must leave r3 alone, it is return value for memcpy etc
@@ -96,9 +96,9 @@ Lmemmove_g3:						// void* memmove(void *dst, const void *src, size_t len)
     // Long operands (more than 32 bytes.)
     //		w1  = (rd-rs), used to check for alignment
 
-LLong0:								// enter from bcopy()
+LLong0:							// enter from bcopy()
     mr		rs,r3				// must leave r3 alone (it is return value for memcpy)
-LLong1:								// enter from memcpy() and memmove()
+LLong1:							// enter from memcpy() and memmove()
     cmplw	cr1,w1,rc			// set cr1 blt iff we must move reverse
     rlwinm	r0,w1,0,0x3			// are operands relatively word-aligned?
     neg		w2,rd				// prepare to align destination
@@ -144,7 +144,7 @@ LLong1:								// enter from memcpy() and memmove()
     stw		w8,28(rd)
     addi	rd,rd,32
     bdnz	1b
-2:									// rc = remaining bytes (0-31)
+2:								// rc = remaining bytes (0-31)
     mtxer	rc					// set up count for string ops
     mr		r0,rd				// move dest ptr out of the way
     lswx	r5,0,rs				// load xer bytes into r5-r12 (rs==r4)
@@ -170,7 +170,7 @@ LLongFloat:
     stswx	w1,0,rd
     add		rd,rd,w4
     beq-	2f					// pathologic case, no chunks to xfer
-1:									// loop over 32-byte chunks
+1:								// loop over 32-byte chunks
     lfd		f0,0(rs)
     lfd		f1,8(rs)
     lfd		f2,16(rs)
@@ -182,7 +182,7 @@ LLongFloat:
     stfd	f3,24(rd)
     addi	rd,rd,32
     bdnz	1b
-2:									// rc = remaining bytes (0-31)
+2:								// rc = remaining bytes (0-31)
     mtxer	rc					// set up count for string ops
     mr		r0,rd				// move dest ptr out of the way
     lswx	r5,0,rs				// load xer bytes into r5-r12 (rs==r4)
@@ -256,7 +256,7 @@ LReverseFloat:
     stfd	f2,-24(rd)
     stfdu	f3,-32(rd)
     bdnz	1b
-2:									// rc = remaining bytes (0-31)
+2:								// rc = remaining bytes (0-31)
     mtxer	rc					// set up count for string ops
     sub		r4,rs,rc			// point to 1st (leftmost) leftover byte (0..31)
     sub		r0,rd,rc			// move dest ptr out of way

library/cpu/generic/strcat.S

@@ -0,0 +1,16 @@
+#include "../4xx/ppc4xx.inc"
+
+function_prolog(__strcat_ppc)
+	addi	r5,r3,-1
+	addi	r4,r4,-1
+1:	lbzu	r0,1(r5)
+	cmpwi	0,r0,0
+	bne	1b
+	addi	r5,r5,-1
+1:	lbzu	r0,1(r4)
+	cmpwi	0,r0,0
+	stbu	r0,1(r5)
+	bne	1b
+	blr
+
+function_epilog(__strcat_ppc)

library/cpu/generic/strcpy.S

@@ -0,0 +1,95 @@
+/* Optimized strcpy implementation for PowerPC.
+   Copyright (C) 1997-2022 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+/* See strlen.s for comments on how the end-of-string testing works.  */
+/* char * [r3] strcpy (char *dest [r3], const char *src [r4])  */
+#include "../4xx/ppc4xx.inc"
+function_prolog(__strcpy_ppc)
+
+#define L(x) 	.L##x
+#define rTMP	r0
+#define rRTN	r3	/* incoming DEST arg preserved as result */
+#define rSRC	r4	/* pointer to previous word in src */
+#define rDEST	r5	/* pointer to previous word in dest */
+#define rWORD	r6	/* current word from src */
+#define rFEFE	r7	/* constant 0xfefefeff (-0x01010101) */
+#define r7F7F	r8	/* constant 0x7f7f7f7f */
+#define rNEG	r9	/* ~(word in s1 | 0x7f7f7f7f) */
+#define rALT	r10	/* alternate word from src */
+
+	or		rTMP, rSRC, rRTN
+	clrlwi.	rTMP, rTMP, 30
+	addi	rDEST, rRTN, -4
+	bne		L(unaligned)
+	lis		rFEFE, -0x101
+	lis		r7F7F, 0x7f7f
+	lwz		rWORD, 0(rSRC)
+	addi	rFEFE, rFEFE, -0x101
+	addi	r7F7F, r7F7F, 0x7f7f
+	b		L(g2)
+L(g0):
+	lwzu	rALT, 4(rSRC)
+	stwu	rWORD, 4(rDEST)
+	add		rTMP, rFEFE, rALT
+	nor		rNEG, r7F7F, rALT
+	and.	rTMP, rTMP, rNEG
+	bne-	L(g1)
+	lwzu	rWORD, 4(rSRC)
+	stwu	rALT, 4(rDEST)
+L(g2):
+	add		rTMP, rFEFE, rWORD
+	nor		rNEG, r7F7F, rWORD
+	and.	rTMP, rTMP, rNEG
+	beq+	L(g0)
+	mr		rALT, rWORD
+/* We've hit the end of the string.  Do the rest byte-by-byte.  */
+L(g1):
+	rlwinm.	rTMP, rALT, 8, 24, 31
+	stb		rTMP, 4(rDEST)
+	beqlr-
+	rlwinm.	rTMP, rALT, 16, 24, 31
+	stb		rTMP, 5(rDEST)
+	beqlr-
+	rlwinm.	rTMP, rALT, 24, 24, 31
+	stb		rTMP, 6(rDEST)
+	beqlr-
+	stb		rALT, 7(rDEST)
+	blr
+/* Oh well.  In this case, we just do a byte-by-byte copy.  */
+	.align 4
+	nop
+L(unaligned):
+	lbz		rWORD, 0(rSRC)
+	addi	rDEST, rRTN, -1
+	cmpwi	rWORD, 0
+	beq-	L(u2)
+L(u0):
+	lbzu	rALT, 1(rSRC)
+	stbu	rWORD, 1(rDEST)
+	cmpwi	rALT, 0
+	beq-	L(u1)
+	nop		/* Let 601 load start of loop.  */
+	lbzu	rWORD, 1(rSRC)
+	stbu	rALT, 1(rDEST)
+	cmpwi	rWORD, 0
+	bne+	L(u0)
+L(u2):
+	stb	rWORD, 1(rDEST)
+	blr
+L(u1):
+	stb	rALT, 1(rDEST)
+	blr
+
+function_epilog(__strcpy_ppc)