Merge pull request #166 from davidgiven/dtrg-comparisons

i80: grabbag of optimisations

mach/i80/libem/cmps_mag.s

@@ -0,0 +1,19 @@
+.sect .text
+.sect .rom
+.sect .data
+.sect .bss
+.sect .text
+
+! Does a tristate signed comparison of hl <> de.
+! Returns m flag if hl < de.
+! Returns p flag if hl >= de.
+! This doesn't set z coherently.
+
+.define .cmps_mag
+.cmps_mag:
+    mov a, d
+    xra h
+    jp .cmpu_mag ! signs are the same, so an unsigned comparison will do
+    xra h        ! set A=H and set the sign flag
+    ral          ! move sign flag into carry
+    ret

mach/i80/libem/cmpu_mag.s

@@ -0,0 +1,18 @@
+.sect .text
+.sect .rom
+.sect .data
+.sect .bss
+.sect .text
+
+! Does a tristate unsigned comparison of hl <> de.
+! Returns m flag if hl < de.
+! Returns p flag if hl >= de.
+! This doesn't set z coherently.
+
+.define .cmpu_mag
+.cmpu_mag:
+    mov a, e
+    sub l
+    mov a, d
+    sbb h
+    ret

mach/i80/libem/rst.s

@@ -11,6 +11,9 @@
 !     41 call .floadn4
 !     34 call .fload4
 !     28 call .fstoren2
+!
+! Also:
+!     48 call .cmps_mag
 
 .define .rst_init
 .rst_init:
@@ -18,13 +21,12 @@
     lxi d, 0x0008
     call copy
     lxi h, .floadn4
-    lxi d, 0x0010
     call copy
     lxi h, .fload4
-    lxi d, 0x0018
     call copy
     lxi h, .fstoren2
-    lxi d, 0x0020
+    call copy
+    lxi h, .cmps_mag
     jmp copy
 
 ! Copies eight bytes from HL to DE.
@@ -34,7 +36,7 @@ copy:
     mov a, m
     stax d
     inx h
-    inx d
+    inr e
     dcr c
     jnz .1
     ret

mach/i80/ncg/table

@@ -158,42 +158,61 @@ INSTRUCTIONS
 MOVES
 
    from reg to reg
-gen mov %2,%1
+      gen
+         mov %2,%1
 
    from const1 %num==0 to areg
-gen xra a
+      gen
+         xra a
 
    from const1 to reg
-gen mvi %2,%1
+      gen
+         mvi %2,%1
 
    from immediate to b_d_h_sp
-gen lxi %2,%1
+      gen
+         lxi %2,%1
+
+   from const1 to reg
+      gen
+         mvi %2, %1
+         
+   from const1 to regpair
+      gen
+         mvi %2.2, %1
+         mvi %2.1, {const1, 0}
 
    from reg to regpair
-gen mov %2.2,%1
+      gen
+         mov %2.2, %1
          mvi %2.1, {const1,0}
 
    from regpair to regpair
-gen mov %2.1,%1.1
+      gen
+         mov %2.1, %1.1
          mov %2.2, %1.2
 
 TESTS
 
    to test areg		/* dummy test, never used */
-gen ora a
+      gen
+         ora a
 
 STACKINGRULES
 
    from regpair to STACK
-gen push %1
+      gen
+         push %1
 
    from immediate + reg to STACK
       uses hl_or_de
-gen move %1,%a
+      gen
+         move %1, %a
          push %a
 
    from immediate + reg to STACK
-gen push hl
+      gen
+         push hl
          move %1, hl
          xthl.
 
@@ -201,34 +220,44 @@ COERCIONS
 
    from STACK
       uses regpair
-gen pop %a				yields %a
+      gen
+         pop %a
+      yields %a
 
    from STACK
       uses hl_or_de
-gen pop %a				yields %a.2
+      gen
+         pop %a
+      yields %a.2
 
    from STACK
       uses areg
-gen dcx sp
+      gen
+         dcx sp
          pop psw
-    inx sp				yields a
+         inx sp
+      yields %a
 
    from immediate
-uses regpair
+      uses regpair=%1
-gen move %1,%a				yields %a
+      yields %a
 
    from hl_or_de
       uses hl_or_de
-gen xchg.				yields %a
+      gen
+         xchg.
+      yields %a
 
    from regpair
-uses regpair
+      uses regpair=%1
-gen move %1,%a				yields %a
+      yields %a
 
    from reg
       uses reusing %1, hl_or_de
-gen move %1,%a.2
+      gen
-    move {const1,0},%a.1		yields %a
+         move %1,%a.2
+         move {const1,0},%a.1
+      yields %a
 
    from hl_or_de				yields %1.2
 
@@ -239,12 +268,36 @@ from smallnconst2
       yields {const2, %1.num}
 
    from const2
-uses hl_or_de
+      uses hl_or_de=%1
-gen move %1,%a				yields %a.2
+      yields %a
+
+   from smallpconst2
+      uses reg={const1, %1.num & 0xff}
+      yields %a
+
+   from smallnconst2
+      uses reg={const1, %1.num & 0xff}
+      yields %a
+
+   from smallpconst2 %1.num == 1
+      uses reg={const1, 0}
+      gen
+         inr %a
+      yields %a
+
+   from smallnconst2 %1.num == 0-1
+      uses reg={const1, 0}
+      gen
+         dcr %a
+      yields %a
+
+   from const2
+      uses reg={const1, %1.num & 0xff}
+      yields %a
 
    from hl_or_de
-uses areg
+      uses reg=%1.2
-gen move %1.2,a				yields a
+      yields %a
 
 PATTERNS
 
@@ -570,12 +623,14 @@ pat stf
 
 pat sti $1==1
    with label areg
-   gen sta %1
+      gen
+         sta %1
    with dereg areg
-   gen stax de
+      gen
+         stax %1
    with hlreg reg
-   gen mov {m},%2
+      gen
-
+         mov {m}, %2
 
 pat sti $1==2
    with label hlreg
@@ -1618,6 +1673,70 @@ leaving cal ".cmf4" asp 8 lfr 2
 pat cmf $1==8
 leaving cal ".cmf8" asp 16 lfr 2
 
+pat cmu zlt $1==2
+   with hl_or_de hl_or_de STACK
+      uses areg
+      gen
+         mov a, %2.2
+         sub %1.2
+         mov a, %2.1
+         sbb %1.1
+         jc {label, $2}
+   with const2 hl_or_de STACK
+      uses areg
+      gen
+         mov a, %2.2
+         sbi {const1, %1.num & 0xff}
+         mov a, %2.1
+         sbi {const1, %1.num >> 8}
+         jc {label, $2}
+   with hl_or_de const2 STACK
+      uses areg
+      gen
+         mvi a, {const1, %2.num & 0xff}
+         sub %1.2
+         mvi a, {const1, %2.num >> 8}
+         sbb %1.1
+         jc {label, $2}
+
+pat cmu zgt $1==2
+   leaving
+      exg 2
+      cmu 2
+      zlt $2
+
+pat cmu zge $1==2
+   with hl_or_de hl_or_de STACK
+      uses areg
+      gen
+         mov a, %2.2
+         sub %1.2
+         mov a, %2.1
+         sbb %1.1
+         jnc {label, $2}
+   with const2 hl_or_de STACK
+      uses areg
+      gen
+         mov a, %2.2
+         sbi {const1, %1.num & 0xff}
+         mov a, %2.1
+         sbi {const1, %1.num >> 8}
+         jnc {label, $2}
+   with hl_or_de const2 STACK
+      uses areg
+      gen
+         mvi a, {const1, %2.num & 0xff}
+         sub %1.2
+         mvi a, {const1, %2.num >> 8}
+         sbb %1.1
+         jnc {label, $2}
+
+pat cmu zle $1==2
+   leaving
+      exg 2
+      cmu 2
+      zge $2
+
 pat cmu $1==2
 with hl_or_de hl_or_de
 uses areg
@@ -1875,41 +1994,14 @@ pat bra
          jmp {label,$1}
 
 pat blt
-   with const2 hl_or_de STACK
+   with hlreg dereg STACK
       uses areg
       gen
-         mov a, %2.1
+         #ifdef USE_I80_RSTS
-         xri {const1, 0x80}
+            rst {const1, 5}
-         mov %2.1, a
+         #else
-         mov a, %2.2
+            Call {label, ".cmps_mag"}
-         sui {const1, %1.num & 0xff}
+         #endif
-         mov a, %2.1
-         sbi {const1, (%1.num >> 8) ^ 0x80}
-         jc {label, $1}
-   with hl_or_de const2 STACK
-      uses areg
-      gen
-         mov a, %1.1
-         xri {const1, 0x80}
-         mov %1.1, a
-         mvi a, {const1, %2.num & 0xff}
-         sub %1.2
-         mvi a, {const1, (%2.num >> 8) ^ 0x80}
-         sbb %1.1
-         jc {label, $1}
-   with hl_or_de hl_or_de STACK
-      uses areg
-      gen
-         mov a, %2.1
-         xri {const1, 0x80}
-         mov %2.1, a
-         mov a, %1.1
-         xri {const1, 0x80}
-         mov %1.1, a
-         mov a, %2.2
-         sub %1.2
-         mov a, %2.1
-         sbb %1.1
          jc {label, $1}
 
 pat bgt
@@ -1918,41 +2010,14 @@ pat bgt
       blt $1
 
 pat bge
-   with const2 hl_or_de STACK
+   with hlreg dereg STACK
       uses areg
       gen
-         mov a, %2.1
+         #ifdef USE_I80_RSTS
-         xri {const1, 0x80}
+            rst {const1, 5}
-         mov %2.1, a
+         #else
-         mov a, %2.2
+            Call {label, ".cmps_mag"}
-         sui {const1, %1.num & 0xff}
+         #endif
-         mov a, %2.1
-         sbi {const1, (%1.num >> 8) ^ 0x80}
-         jnc {label, $1}
-   with hl_or_de const2 STACK
-      uses areg
-      gen
-         mov a, %1.1
-         xri {const1, 0x80}
-         mov %1.1, a
-         mvi a, {const1, %2.num & 0xff}
-         sub %1.2
-         mvi a, {const1, (%2.num >> 8) ^ 0x80}
-         sbb %1.1
-         jnc {label, $1}
-   with hl_or_de hl_or_de STACK
-      uses areg
-      gen
-         mov a, %2.1
-         xri {const1, 0x80}
-         mov %2.1, a
-         mov a, %1.1
-         xri {const1, 0x80}
-         mov %1.1, a
-         mov a,%2.2
-         sub %1.2
-         mov a,%2.1
-         sbb %1.1
          jnc {label, $1}
 
 pat ble
@@ -2012,59 +2077,65 @@ pat bne
 
 pat zlt
    with STACK
-gen pop psw
+      gen
+         pop psw
          ora a
          jm {label,$1}
    with hl_or_de STACK
-gen mov a,%1.1
+      gen
+         mov a,%1.1
          ora a
          jm {label,$1}
 
 pat zle
    with hl_or_de STACK
       uses areg
-gen xra a
+      gen
-    add %1.1
+         mov a, %1.1
+         ora a
          jm {label, $1}
          jnz {label, 1f}
-    xra a
+         ora %1.2
-    add %1.2
          jz {label, $1}
          1:
 
 pat zeq
    with hl_or_de STACK
       uses areg
-gen mov a,%1.1
+      gen
+         mov a,%1.1
          ora %1.2
          jz {label,$1}
 
 pat zne
    with hl_or_de STACK
       uses areg
-gen mov a,%1.1
+      gen
+         mov a,%1.1
          ora %1.2
          jnz {label,$1}
 
 pat zge
    with STACK
-gen pop psw
+      gen
+         pop psw
          ral.
          jnc {label,$1}
    with hl_or_de STACK
-gen mov a,%1.1
+      gen
+         mov a,%1.1
          ora a
          jp {label,$1}
 
 pat zgt
    with hl_or_de STACK
       uses areg
-gen xra a
+      gen
-    add %1.1
+         mov a, %1.1
+         ora a
          jm {label, 1f}
          jnz {label, $1}
-    xra a
+         ora %1.2
-    add %1.2
          jnz {label, $1}
          1:
 
@@ -2091,7 +2162,8 @@ pat lol zne
 pat ior zeq $1==2
    with hl_or_de hl_or_de STACK
       uses areg
-gen mov a,%1.1
+      gen
+         mov a,%1.1
          ora %1.2
          ora %2.1
          ora %2.2
@@ -2100,7 +2172,8 @@ gen mov a,%1.1
 pat ior zne $1==2
    with hl_or_de hl_or_de STACK
       uses areg
-gen mov a,%1.1
+      gen
+         mov a,%1.1
          ora %1.2
          ora %2.1
          ora %2.2
@@ -2165,14 +2238,14 @@ gen 1:
 pat asp $1==0 /* do nothing */
 
 pat asp ($1==2)
-   with regpair
+   with hl_or_de
    with STACK
       uses hlreg
       gen
          pop hl
 
 pat asp ($1==4)
-   with regpair regpair
+   with hl_or_de hl_or_de
    with STACK
       uses hlreg
       gen

mach/i80/top/table

@@ -16,4 +16,9 @@ xchg : inx d : xchg          -> inx h ;
 cpi 0                        -> ora a ;
 call X : ret                 -> jmp X ;
 
+push h : lxi h, X : pop d    -> lxi d, X : xchg ;
+push d : lxi d, X : pop h    -> lxi h, X : xchg ;
+
+push h : lhld h, X : pop d    -> xchg : lhld X ;
+
 %%;