x86 assembler fixes:

* mkx86ins.pp: 'regmem' operand type in x86ins.dat must be converted to OT_RM_GPR, not OT_REGMEM. The latter does not restrict register type, allowing to use e.g. xmm registers in place of regular ones. 
* Finished 'movd' and 'movq', added some tests for them to tasm2.pp.

git-svn-id: trunk@17515 -

compiler/i386/i386nop.inc

@@ -1,2 +1,2 @@
 { don't edit, this file is generated from x86ins.dat }
-1210;
+1202;

compiler/utils/mkx86ins.pp

@@ -59,12 +59,12 @@ function formatop(s:string;allowsizeonly:boolean):string;
      replacetab : array[1..replaces,1..2] of string[32]=(
        (':',' or ot_colon'),
        ('reg','regnorm'),
-       ('regmem','regmem'),
-       ('rm8','regmem or ot_bits8'),
-       ('rm16','regmem or ot_bits16'),
-       ('rm32','regmem or ot_bits32'),
-       ('rm64','regmem or ot_bits64'),
-       ('rm80','regmem or ot_bits80'),
+       ('regmem','rm_gpr'),
+       ('rm8','rm_gpr or ot_bits8'),
+       ('rm16','rm_gpr or ot_bits16'),
+       ('rm32','rm_gpr or ot_bits32'),
+       ('rm64','rm_gpr or ot_bits64'),
+       ('rm80','rm_gpr or ot_bits80'),
        ('mem8','memory or ot_bits8'),
        ('mem16','memory or ot_bits16'),
        ('mem32','memory or ot_bits32'),
@@ -445,5 +445,5 @@ begin
    writeln(nopfile,insns,';');
    close(nopfile);
    closeinc(propfile);
-   writeln(insns,' nodes procesed (maxinfolen=',maxinfolen,')');
+   writeln(insns,' nodes processed (maxinfolen=',maxinfolen,')');
 end.

compiler/x86/x86ins.dat

@@ -1063,23 +1063,21 @@ reg_treg,reg64        \2\x0F\x26\110                  386,PRIV,X86_64
 
 [MOVD,movd]
 (Ch_Rop1, Ch_Wop2, Ch_None)
-mmxreg,mem            \2\x0F\x6E\110                  PENT,MMX,SD
-mmxreg,reg32          \2\x0F\x6E\110                  PENT,MMX
-mem,mmxreg            \2\x0F\x7E\101                  PENT,MMX,SD
-reg32,mmxreg          \2\x0F\x7E\101                  PENT,MMX
-xmmreg,reg32          \361\2\x0F\x6E\110              WILLAMETTE,SSE2
-reg32,xmmreg          \361\2\x0F\x7E\101              WILLAMETTE,SSE2
-mem,xmmreg            \361\325\2\x0F\x7E\101          WILLAMETTE,SSE2
-xmmreg,mem            \361\325\2\x0F\x6E\110          WILLAMETTE,SSE2
+mmxreg,rm32           \2\x0F\x6E\110                  PENT,MMX,SD
+rm32,mmxreg           \2\x0F\x7E\101                  PENT,MMX,SD
+xmmreg,rm32           \361\2\x0F\x6E\110              WILLAMETTE,SSE2
+rm32,xmmreg           \361\2\x0F\x7E\101              WILLAMETTE,SSE2
 
 [MOVQ,movq]
 (Ch_Rop1, Ch_Wop2, Ch_None)
 mmxreg,mmxrm          \2\x0F\x6F\110                  PENT,MMX,SM
 mmxrm,mmxreg          \2\x0F\x7F\101                  PENT,MMX,SM
+mmxreg,rm64           \326\2\x0F\x6E\110              X86_64,MMX
+rm64,mmxreg           \326\2\x0F\x7E\101              X86_64,MMX
 xmmreg,xmmrm          \333\2\x0F\x7E\110              WILLAMETTE,SSE2
 xmmrm,xmmreg          \361\2\x0F\xD6\101              WILLAMETTE,SSE2
-xmmreg,reg64          \361\326\2\x0F\x6E\110          WILLAMETTE,SSE2
-reg64,xmmreg          \361\326\2\x0F\x7E\101          WILLAMETTE,SSE2
+xmmreg,reg64          \361\326\2\x0F\x6E\110          WILLAMETTE,SSE2,X86_64
+reg64,xmmreg          \361\326\2\x0F\x7E\101          WILLAMETTE,SSE2,X86_64
 
 [MOVSB]
 (Ch_All, Ch_None, Ch_None)
@@ -1089,10 +1087,8 @@ void                  \1\xA4                          8086
 ; Ch_All isn't correct for the sse move, but how can it be solved? (FK)
 (Ch_All, Ch_None, Ch_None)
 void                  \325\1\xA5                      386
-xmmreg,xmmreg         \334\2\x0F\x10\110              WILLAMETTE,SSE2
-xmmreg,xmmreg         \334\2\x0F\x11\110              WILLAMETTE,SSE2
-mem,xmmreg            \334\2\x0F\x11\101              WILLAMETTE,SSE2
-xmmreg,mem            \334\2\x0F\x10\110              WILLAMETTE,SSE2
+xmmreg,xmmrm          \334\2\x0F\x10\110              WILLAMETTE,SSE2
+xmmrm,xmmreg          \334\2\x0F\x11\101              WILLAMETTE,SSE2
 
 [MOVSQ]
 (Ch_All, Ch_None, Ch_None)
@@ -3029,10 +3025,6 @@ mem                   \326\2\x0F\xC7\201              X86_64
 ;
 ; SSE4a (AMD Barcelona CPUs, n/a on Intel)
 ;
-
-
-; note: \333=F3h, \334=F2h, \336=66h, \325=no REX.W=1 for qword, \375=unsigned
-
 [MOVNTSS]
 (Ch_All, Ch_None, Ch_None)
 mem,xmmreg              \333\2\x0F\x2B\101                        SSE4,SD
@@ -3045,11 +3037,11 @@ mem,xmmreg              \334\325\2\x0F\x2B\101                SSE4 ;,SQ
 (Ch_All, Ch_None, Ch_None)
 xmmreg,xmmreg           \334\2\x0F\x79\110                        SSE4
 ; four operands are not possible yet
-; xmmreg,xmmreg,imm,imm   \110\334\76\2\x0F\x78\77\375\22\375\23          SSE4,SB
+; xmmreg,xmmreg,imm,imm   \334\2\x0F\x78\110\26\27          SSE4,SB
 
 [EXTRQ]
 (Ch_All, Ch_None, Ch_None)
-xmmreg,imm,imm          \361\2\x0F\x78\200\375\21\375\22          SSE4,SB
+xmmreg,imm,imm          \361\2\x0F\x78\200\25\26                  SSE4,SB
 xmmreg,xmmreg           \361\2\x0F\x79\110                        SSE4
 
 [LZCNT,lzcntX]

compiler/x86_64/x8664nop.inc

@@ -1,2 +1,2 @@
 { don't edit, this file is generated from x86ins.dat }
-1203;
+1199;

tests/test/tasm2.pp

@@ -156,11 +156,39 @@ asm
     psadbw (%r8), %xmm0
 end;
 
+procedure tmovd; assembler; nostackframe;
+asm
+    movd  0x12345678(%rip), %xmm0
+    movd  %xmm0, 0x12345678(%rip)
+    movd  %eax, %xmm0
+    movd  %xmm0, %eax
+
+{ same for MMX registers }
+    movd  0x12345678(%rip), %mm0
+    movd  %mm0, 0x12345678(%rip)
+    movd  %eax, %mm0
+    movd  %mm0, %eax
+end;
+
+procedure tmovq; assembler; nostackframe;
+asm
+    movq  0x12345678(%rip), %xmm0
+    movq  %xmm0, 0x12345678(%rip)
+    movq  %xmm1, %xmm0
+    movq  %rax, %xmm0
+    movq  %xmm0, %rax
+
+{ same for MMX registers }
+    movq  0x12345678(%rip), %mm0
+    movq  %mm0, 0x12345678(%rip)
+    movq  %mm1, %mm0
+    movq  %rax, %mm0
+    movq  %mm0, %rax
+end;
+
 // Here are some more tough nuts to crack...
 procedure test2; assembler; nostackframe;
 asm
-//    movq      %xmm1,0x12345678(%rip)    { converted to mov, invalid combination of opcode and operands }
-//    movq      0x12345678(%rip),%xmm1
 //    cmpsd     $0x10,%xmm7,%xmm6         { unrecognized opcode }
 //    cmpsd     $0x10,(%rax),%xmm7
 //    movsd     (%rax),%xmm8              { unrecognized opcode }
@@ -221,16 +249,46 @@ const
     $58,$00,$F3,$41,$0F,$58,$00,$66,$41,$0F,$E4,$00,$66,$41,$0F,
     $F6,$00,$C3);
 
-var
-  i : longint;
+ tmovq_expected: array[0..54] of byte = (
+    $F3,$0F,$7E,$05,$78,$56,$34,$12,
+    $66,$0F,$D6,$05,$78,$56,$34,$12,
+    $F3,$0F,$7E,$C1,
+    $66,$48,$0F,$6E,$C0,
+    $66,$48,$0F,$7E,$C0,
+    $0F,$6F,$05,$78,$56,$34,$12,
+    $0F,$7F,$05,$78,$56,$34,$12,
+    $0F,$6F,$C1,
+    $48,$0F,$6E,$C0,
+    $48,$0F,$7E,$C0
+  );
 
+  tmovd_expected: array[0..43] of byte = (
+    $66,$0F,$6E,$05,$78,$56,$34,$12,
+    $66,$0F,$7E,$05,$78,$56,$34,$12,
+    $66,$0F,$6E,$C0,
+    $66,$0F,$7E,$C0,
+    $0F,$6E,$05,$78,$56,$34,$12,
+    $0F,$7E,$05,$78,$56,$34,$12,
+    $0F,$6E,$C0,
+    $0F,$7E,$C0
+  );
 
+
+procedure check(const id: string; const expected: array of byte; p: pointer);
+var
+  i : longint;
 begin
-  for i:=0 to high(test_expected) do
-    if test_expected[i]<>pbyte(@test)[i] then
+  for i:=0 to high(expected) do
+    if expected[i]<>pbyte(p)[i] then
       begin
-        writeln('mismatch at offset $',hexstr(i,4), ', expected=$',hexstr(test_expected[i],2),' actual=$',hexstr(pbyte(@test)[i],2));
+        writeln(id, ' mismatch at offset $',hexstr(i,4), ', expected=$',hexstr(expected[i],2),' actual=$',hexstr(pbyte(p)[i],2));
         halt(1);
       end;
+end;
+
+begin
+  check('generic', test_expected, @test);
+  check('movq', tmovq_expected, @tmovq);
+  check('movd', tmovd_expected, @tmovd);
   writeln('ok');
 end.