diff --git a/rtl/powerpc/set.inc b/rtl/powerpc/set.inc index d2ace9af64..e15bad399a 100644 --- a/rtl/powerpc/set.inc +++ b/rtl/powerpc/set.inc @@ -22,7 +22,7 @@ function fpc_set_load_small(l: fpc_small_set): fpc_normal_set;assembler;[public, on entry: p in r3, l in r4 } asm - stw r4,(r3) + stw r4,0(r3) li r0,0 stw r0,4(r3) stw r0,8(r3) @@ -34,6 +34,7 @@ asm end ['r0']; +{ checked 2001/09/28 (JM) } function fpc_set_create_element(b : byte): fpc_normal_set;assembler;[public,alias:'FPC_SET_CREATE_ELEMENT']; compilerproc; { create a new set in p from an element b @@ -42,7 +43,7 @@ function fpc_set_create_element(b : byte): fpc_normal_set;assembler;[public,alia } asm li r0,0 - stw r0,(r3) + stw r0,0(r3) stw r0,4(r3) stw r0,8(r3) stw r0,12(r3) @@ -51,15 +52,15 @@ asm stw r0,24(r3) stw r0,28(r3) - // r0 := 1 shl r4[27-31] -> bit index in dword (shift instructions + // r0 := 1 shl r4[27-31] -> bit index in dword (rotate instructions // with count in register only consider lower 5 bits of this register) li r0,1 - slw r0,r0,r4 + rlwnm r0,r0,r4,0,31 // get the index of the correct *dword* in the set // (((b div 8) div 4)*4= (b div 8) and not(3)) - // r5 := (r4 rotl(32-3)) and (0x0fffffff8) - rlwinm r4,r4,29,0,31-2 + // r5 := (r4 rotl(32-3)) and (0x01ffffff8) + rlwinm r4,r4,31-3+1,3,31-2 // store the result stwx r0,r3,r4 @@ -84,12 +85,13 @@ Lset_set_byte_copy: subi r3,r3,32 // get the index of the correct *dword* in the set // r0 := (r5 rotl(32-3)) and (0x0fffffff8) - rlwinm r0,r5,29,0,31-2 + rlwinm r0,r5,31-3+1,3,31-2 // load dword in which the bit has to be set (and update r3 to this address) lwzxu r4,r3,r0 li r0,1 // generate bit which has to be inserted - slw r5,r0,r5 + // (can't use rlwimi, since that one only works for constants) + rlwnm r5,r0,r5 // insert it or r5,r4,r5 // store result @@ -117,12 +119,12 @@ Lset_unset_byte_copy: subi r3,r3,32 // get the index of the correct *dword* in the set // r0 := (r4 rotl(32-3)) and (0x0fffffff8) - rlwinm r0,r5,29,0,31-2 + rlwinm r0,r5,31-3+1,3,31-2 // load dword in which the bit has to be set (and update r3 to this address) lwzxu r4,r3,r0 li r0,1 // generate bit which has to be removed - slw r5,r0,r5 + rlwnm r5,r0,r5,0,31 // remove it andc r5,r4,r5 // store result @@ -149,16 +151,16 @@ Lset_set_range_copy: subi r3,r3,32 cmplw cr0,r5,r6 - bg cr0,LSET_RANGE_EXIT - rlwinm r4,r5,32-3,0,31-2 // divide by 8 to get starting and ending byte- + bg cr0,Lset_range_exit + rlwinm r4,r5,31-3+1,3,31-2 // divide by 8 to get starting and ending byte- { load the set the data cache } dcbst r3,r4 - rlwinm r9,r5,32-3,0,31-2 // address and clear two lowest bits to get + rlwinm r9,r6,31-3+1,3,31-2 // address and clear two lowest bits to get // start/end longint address sub. r9,r4,r9 // are bit lo and hi in the same longint? - rlwinm r6,r6,0,31-4,31 // hi := hi mod 32 (= "hi and 31", but the andi + rlwinm r6,r6,0,31-5+1,31 // hi := hi mod 32 (= "hi and 31", but the andi // instr. only exists in flags modifying form) - li r10,$ffff // r10 = $0x0ffffffff = bitmask to be inserted + li r10,-1 // r10 = $0x0ffffffff = bitmask to be inserted subfic r6,r6,31 // hi := 31 - (hi mod 32) = shift count for later srw r10,r10,r4 // shift bitmask to clear bits below lo // note: shift right = opposite little endian!! @@ -169,7 +171,7 @@ Lset_set_range_copy: subic. r9,r9,4 // bit hi in next longint? or r5,r5,r10 // merge and stw r5,(r3) // store current mask - li r10,$ffff // new mask + li r10,-1 // new mask lwzu r5,4(r3) // load next longint of set beq Lset_range_hi // bit hi in this longint -> go to adjust for hi Lset_range_loop: @@ -200,12 +202,12 @@ function fpc_set_in_byte(const p: fpc_normal_set; b : byte): boolean;assembler;[ asm // get the index of the correct *dword* in the set // r0 := (r4 rotl(32-3)) and (0x0fffffff8) - rlwinm r0,r4,29,0,31-2 + rlwinm r0,r4,31-3+1,3,31-2 // load dword in which the bit has to be tested lwzx r3,r3,r0 li r0,1 // generate bit which has to be tested - slw r4,r0,r4 + rwlwnm r4,r0,r4,0,31 // test it and. r3,r3,r4 end ['r0','r3','r4','cr0']; @@ -507,7 +509,14 @@ end; { $Log$ - Revision 1.9 2001-09-27 15:30:29 jonas + Revision 1.10 2001-09-28 13:27:02 jonas + * use rlwnm instead of slw, because, although the programming + environments manual states otherwise, slw uses the whole contents of + the register instead of bits 27-31 as shift count (rlwnm doesn't) + * fixed generation of offset inside normal sets where bits have to be + inserted + + Revision 1.9 2001/09/27 15:30:29 jonas * conversion to compilerproc and to structure used by i386 rtl * some bugfixes * powerpc.inc is almost complete (only fillchar/word/dword, get_frame etc