Added pce-specific substitutes for memcpy() and memmove().

They are smaller and faster because they take advantage of the pce CPU's block-copy instructions.

Also, made a small improvement to the common memmove(), so that it is similar to the pce version.

libsrc/common/memmove.s

@@ -1,7 +1,7 @@
 ;
-; Ullrich von Bassewitz, 2003-08-20
-; Performance increase (about 20%) by
-; Christian Krueger, 2009-09-13
+; 2003-08-20, Ullrich von Bassewitz
+; 2009-09-13, Christian Krueger -- performance increase (about 20%)
+; 2015-10-23, Greg King
 ;
 ; void* __fastcall__ memmove (void* dest, const void* src, size_t size);
 ;
@@ -23,8 +23,7 @@ _memmove:
 ; low addresses and increase pointers), otherwise we must copy downwards
 ; (start at high addresses and decrease pointers).
 
-        sec
-        sbc     ptr1
+        cmp     ptr1
         txa
         sbc     ptr1+1
         jcc     memcpy_upwards  ; Branch if dest < src (upwards copy)
@@ -81,4 +80,3 @@ PageSizeCopy:                   ; assert Y = 0
 ; Done, return dest
 
 done:   jmp     popax           ; Pop ptr and return as result
-                

libsrc/pce/memcpy.s

@@ -0,0 +1,94 @@
+;
+; This file, instead of "common/memcpy.s", will be assembled for the pce
+; target.  This version is smaller and faster because it uses the HuC6280's
+; block-copy instructions.
+;
+; 2003-08-20, Ullrich von Bassewitz
+; 2015-10-11, Greg King
+;
+; void* __fastcall__ memcpy (void* dest, const void* src, size_t size);
+;
+; NOTE:  This function contains entry points for memmove, which will resort
+; to memcpy for an incrementing copy.  Don't change this module without looking
+; at "pce/memmove.s"!
+;
+
+        .export         _memcpy
+        .export         memcpy_increment, memcpy_transfer, memcpy_getparams
+
+        .import         popax
+        .importzp       sp, ptr1, ptr2, ptr3
+
+
+; The structure of the transfer instructions
+
+        .struct
+opcode          .byte
+source          .addr
+destination     .addr
+length          .word
+        .endstruct
+
+; ----------------------------------------------------------------------
+_memcpy:
+        jsr     memcpy_getparams
+
+memcpy_increment:
+        ldy     #$73                    ; TII
+
+memcpy_transfer:
+        sty     transfer+opcode
+
+        lda     ptr1
+        ldx     ptr1+1
+        sta     transfer+source
+        stx     transfer+source+1
+
+        lda     ptr2
+        ldx     ptr2+1
+        sta     transfer+destination
+        stx     transfer+destination+1
+
+        lda     ptr3
+        ldx     ptr3+1
+        sta     transfer+length
+        stx     transfer+length+1
+
+        jmp     transfer
+
+; ----------------------------------------------------------------------
+; Get the parameters from the stack, as follows:
+;
+;       size --> ptr3
+;       src  --> ptr1
+;       dest --> ptr2
+;
+; The first argument (dest) will remain on the stack; and, is returned in .XA!
+
+memcpy_getparams:
+        sta     ptr3
+        stx     ptr3+1                  ; save size
+
+        jsr     popax
+        sta     ptr1
+        stx     ptr1+1                  ; save src
+
+; (Direct stack access is four cycles faster [total cycle count].)
+
+        ldy     #1                      ; save dest
+        lda     (sp),y                  ; get high byte
+        tax
+        lda     (sp)                    ; get low byte
+        sta     ptr2
+        stx     ptr2+1
+        rts                             ; return dest address (for memmove)
+
+; ----------------------------------------------------------------------
+; The transfer instructions use inline arguments.
+; Therefore, we must build the instruction in the DATA segment.
+
+.data
+
+transfer:
+        tii     $FFFF, $FFFF, $0001
+        jmp     popax                   ; get pointer; and, return it as result

libsrc/pce/memmove.s

@@ -0,0 +1,63 @@
+;
+; This file, instead of "common/memmove.s", will be assembled for the pce
+; target.  This version is smaller and faster because it uses the HuC6280's
+; block-copy instructions.
+;
+; 2003-08-20, Ullrich von Bassewitz
+; 2015-10-23, Greg King
+;
+; void* __fastcall__ memmove (void* dest, const void* src, size_t size);
+;
+; NOTE: This function uses entry points from "pce/memcpy.s"!
+;
+
+        .export         _memmove
+
+        .import         memcpy_getparams, memcpy_increment, memcpy_transfer
+        .importzp       ptr1, ptr2, ptr3
+
+        .macpack        generic
+        .macpack        longbranch
+
+
+; ----------------------------------------------------------------------
+_memmove:
+        jsr     memcpy_getparams
+
+; Check for the copy direction.  If dest < src, we must copy downwards (start
+; at low addresses, and increase pointers); otherwise, we must copy upwards
+; (start at high addresses, and decrease pointers).
+
+        cmp     ptr1
+        txa
+        sbc     ptr1+1
+        jcc     memcpy_increment        ; Branch if dest < src
+
+; Copy decrementing; adjust the pointers to the end of the memory regions.
+
+        lda     ptr1
+        add     ptr3
+        sta     ptr1
+        lda     ptr1+1
+        adc     ptr3+1
+        sta     ptr1+1
+
+        lda     ptr1                    ; point to last byte of source
+        bne     @L1
+        dec     ptr1+1
+@L1:    dec     ptr1
+
+        lda     ptr2
+        add     ptr3
+        sta     ptr2
+        lda     ptr2+1
+        adc     ptr3+1
+        sta     ptr2+1
+
+        lda     ptr2                    ; point to last byte of target
+        bne     @L2
+        dec     ptr2+1
+@L2:    dec     ptr2
+
+        ldy     #$C3                    ; TDD
+        jmp     memcpy_transfer