Work on floating point support.

5 files changed, 207 insertions, 33 deletions

diff --git a/README.md b/README.md
index 1505a4d..bc9cbe2 100644
--- a/README.md
+++ b/README.md
@@ -29,7 +29,7 @@ cc -O2 *.c -o minias
 
 # Resources
 
-- [intel reference](https://software.intel.com/content/dam/develop/external/us/en/documents-tps/325383-sdm-vol-2abcd.pdf) - Especially chapter 2 section 1.
+- [intel reference](https://software.intel.com/content/dam/develop/external/us/en/documents-tps/325383-sdm-vol-2abcd.pdf) - Specifically chapter 2.1 and chapter 3.1.
 - [elf spec](https://refspecs.linuxfoundation.org/elf/elf.pdf)
 - [goas](https://github.com/DQNEO/goas)
 - [neatas](https://repo.or.cz/neatas.git)
diff --git a/asm.peg b/asm.peg
index 742da11..8923b08 100644
--- a/asm.peg
+++ b/asm.peg
@@ -56,37 +56,42 @@ label =
 
 instr = 
   # XXX Order the rules by frequency to get better lookup time.
-    "nop"   { $$.kind = ASM_NOP; }
-  | "leave" { $$.kind = ASM_LEAVE; }
-  | "ret"   { $$.kind = ASM_RET; }
-  | "cltd"  { $$.kind = ASM_CLTD; }
-  | "cqto"  { $$.kind = ASM_CQTO; }
-  | i:push  { $$ = i; }
-  | i:pop   { $$ = i; }
-  | i:call  { $$ = i; }
-  | i:jmp   { $$ = i; }
-  | i:add   { $$ = i; }
-  | i:and   { $$ = i; }
-  | i:cmp   { $$ = i; }
-  | i:div   { $$ = i; }
-  | i:idiv  { $$ = i; }
-  | i:lea   { $$ = i; }
-  | i:mul   { $$ = i; }
-  | i:imul  { $$ = i; }
-  | i:neg   { $$ = i; }
-  | i:movsx { $$ = i; }
-  | i:movzx { $$ = i; }
-  | i:mov   { $$ = i; }
-  | i:or    { $$ = i; }
-  | i:set   { $$ = i; }
-  | i:sub   { $$ = i; }
-  | i:sal   { $$ = i; }
-  | i:sar   { $$ = i; }
-  | i:shl   { $$ = i; }
-  | i:shr   { $$ = i; }
-  | i:test  { $$ = i; }
-  | i:xchg  { $$ = i; }
-  | i:xor   { $$ = i; }
+    "nop"     { $$.kind = ASM_NOP; }
+  | "leave"   { $$.kind = ASM_LEAVE; }
+  | "ret"     { $$.kind = ASM_RET; }
+  | "cltd"    { $$.kind = ASM_CLTD; }
+  | "cqto"    { $$.kind = ASM_CQTO; }
+  | i:push    { $$ = i; }
+  | i:pop     { $$ = i; }
+  | i:call    { $$ = i; }
+  | i:jmp     { $$ = i; }
+  | i:add     { $$ = i; }
+  | i:and     { $$ = i; }
+  | i:cmp     { $$ = i; }
+  | i:div     { $$ = i; }
+  | i:idiv    { $$ = i; }
+  | i:lea     { $$ = i; }
+  | i:mulsd   { $$ = i; }
+  | i:mulss   { $$ = i; }
+  | i:mul     { $$ = i; }
+  | i:imul    { $$ = i; }
+  | i:neg     { $$ = i; }
+  | i:movsx   { $$ = i; }
+  | i:movzx   { $$ = i; }
+  | i:mov     { $$ = i; }
+  | i:or      { $$ = i; }
+  | i:set     { $$ = i; }
+  | i:sub     { $$ = i; }
+  | i:sal     { $$ = i; }
+  | i:sar     { $$ = i; }
+  | i:shl     { $$ = i; }
+  | i:shr     { $$ = i; }
+  | i:test    { $$ = i; }
+  | i:ucomisd { $$ = i; }
+  | i:ucomiss { $$ = i; }
+  | i:test    { $$ = i; }
+  | i:xchg    { $$ = i; }
+  | i:xor     { $$ = i; }
 
 push = 
   "push" (
@@ -160,6 +165,26 @@ idiv = "idiv" (
   | args:r-opargs { args.instr.variant += 4 ; $$ = args; }
 )  { $$.instr.kind = ASM_IDIV; }
 
+mulsd = "mulsd" (
+    ws s:r-xmm ws? ',' ws? d:r-xmm { $$ = INSTR2(0, s, d); }
+  | ws s:m     ws? ',' ws? d:r-xmm { $$ = INSTR2(1, s, d); }
+)  { $$.instr.kind = ASM_MULSD; }
+
+mulss = "mulss" (
+    ws s:r-xmm ws? ',' ws? d:r-xmm { $$ = INSTR2(0, s, d); }
+  | ws s:m     ws? ',' ws? d:r-xmm { $$ = INSTR2(1, s, d); }
+)  { $$.instr.kind = ASM_MULSS; }
+
+ucomiss = "ucomiss" (
+    ws s:r-xmm ws? ',' ws? d:r-xmm { $$ = INSTR2(0, s, d); }
+  | ws s:m     ws? ',' ws? d:r-xmm { $$ = INSTR2(1, s, d); }
+)  { $$.instr.kind = ASM_UCOMISS; }
+
+ucomisd = "ucomisd" (
+    ws s:r-xmm ws? ',' ws? d:r-xmm { $$ = INSTR2(0, s, d); }
+  | ws s:m     ws? ',' ws? d:r-xmm { $$ = INSTR2(1, s, d); }
+)  { $$.instr.kind = ASM_UCOMISD; }
+
 mul = "mul" (
     args:m-opargs { $$ = args; }
   | args:r-opargs { args.instr.variant += 4 ; $$ = args; }
@@ -440,6 +465,35 @@ r64 =
   | "%r14" ![lwb] { $$ = REG(ASM_R14); }
   | "%r15" ![lwb] { $$ = REG(ASM_R15); }
 
+r-mm = 
+    "%mm0" { $$ = REG(ASM_MM0); }
+  | "%mm1" { $$ = REG(ASM_MM1); }
+  | "%mm2" { $$ = REG(ASM_MM2); }
+  | "%mm3" { $$ = REG(ASM_MM3); }
+  | "%mm4" { $$ = REG(ASM_MM4); }
+  | "%mm5" { $$ = REG(ASM_MM5); }
+  | "%mm6" { $$ = REG(ASM_MM6); }
+  | "%mm7" { $$ = REG(ASM_MM7); }
+
+r-xmm = 
+  # Reverse order due to peg ordering.
+    "%xmm15" { $$ = REG(ASM_XMM15); }
+  | "%xmm14" { $$ = REG(ASM_XMM14); }
+  | "%xmm13" { $$ = REG(ASM_XMM13); }
+  | "%xmm12" { $$ = REG(ASM_XMM12); }
+  | "%xmm11" { $$ = REG(ASM_XMM11); }
+  | "%xmm10" { $$ = REG(ASM_XMM10); }
+  | "%xmm9"  { $$ = REG(ASM_XMM7); }
+  | "%xmm8"  { $$ = REG(ASM_XMM7); }
+  | "%xmm7"  { $$ = REG(ASM_XMM7); }
+  | "%xmm6"  { $$ = REG(ASM_XMM6); }
+  | "%xmm5"  { $$ = REG(ASM_XMM5); }
+  | "%xmm4"  { $$ = REG(ASM_XMM4); }
+  | "%xmm3"  { $$ = REG(ASM_XMM3); }
+  | "%xmm2"  { $$ = REG(ASM_XMM2); }
+  | "%xmm1"  { $$ = REG(ASM_XMM1); }
+  | "%xmm0"  { $$ = REG(ASM_XMM0); }
+
 # We disallow newlines in our strings, it is simpler for lineno tracking.
 string = '"' <(string-escape | ( ! '\n' ! '\\' !'"' .))*> '"'
          { $$.string = decodestring(yytext); }
diff --git a/main.c b/main.c
index a169a42..0d58350 100644
--- a/main.c
+++ b/main.c
@@ -667,6 +667,29 @@ static void assembleshift(Instr *instr, uint8_t immreg) {
   }
 }
 
+static void assemblemuls(Instr *instr, uint8_t prefix) {
+  Opcode opcode;
+
+  opcode = 0x01000f59;
+  sb(prefix);
+
+  if (instr->arg1->kind == ASM_MEMARG) {
+    assemblemem(&instr->arg1->memarg, 0, opcode, regbits(instr->arg2->kind), 8);
+  } else {
+    assemblemodregrm(0, opcode, 3, regbits(instr->arg2->kind),
+                     regbits(instr->arg1->kind), 8);
+  }
+}
+
+static void assembleucomis(Instr *instr, Opcode opcode) {
+  if (instr->arg1->kind == ASM_MEMARG) {
+    assemblemem(&instr->arg1->memarg, 0, opcode, regbits(instr->arg2->kind), 8);
+  } else {
+    assemblemodregrm(0, opcode, 3, regbits(instr->arg2->kind),
+                     regbits(instr->arg1->kind), 8);
+  }
+}
+
 static void assemble(void) {
   Symbol *sym;
   Parsev *v;
@@ -866,6 +889,12 @@ static void assemble(void) {
     case ASM_MUL:
       assembledivmulneg(&v->instr, 0x04);
       break;
+    case ASM_MULSD:
+      assemblemuls(&v->instr, 0xf2);
+      break;
+    case ASM_MULSS:
+      assemblemuls(&v->instr, 0xf3);
+      break;
     case ASM_IMUL: {
       Opcode opcode;
       uint8_t opsz;
@@ -951,6 +980,34 @@ static void assemble(void) {
       }
       break;
     }
+    case ASM_UCOMISD: {
+      Opcode opcode = 0x01000f2e;
+
+      sb(0x66);
+
+      if (v->instr.arg1->kind == ASM_MEMARG) {
+        assemblemem(&v->instr.arg1->memarg, 0, opcode,
+                    regbits(v->instr.arg2->kind), 8);
+      } else {
+        assemblemodregrm(0, opcode, 3, regbits(v->instr.arg2->kind),
+                         regbits(v->instr.arg1->kind), 8);
+      }
+
+      break;
+    }
+    case ASM_UCOMISS: {
+      Opcode opcode = 0x01000f2e;
+
+      if (v->instr.arg1->kind == ASM_MEMARG) {
+        assemblemem(&v->instr.arg1->memarg, 0, opcode,
+                    regbits(v->instr.arg2->kind), 8);
+      } else {
+        assemblemodregrm(0, opcode, 3, regbits(v->instr.arg2->kind),
+                         regbits(v->instr.arg1->kind), 8);
+      }
+
+      break;
+    }
     case ASM_XOR: {
       static uint8_t variant2op[24] = {
           0x34, 0x35, 0x35, 0x35, 0x80, 0x81, 0x81, 0x81,
diff --git a/minias.h b/minias.h
index 750f481..fdcbc09 100644
--- a/minias.h
+++ b/minias.h
@@ -74,6 +74,8 @@ typedef enum {
   ASM_IDIV,
   ASM_LEA,
   ASM_MUL,
+  ASM_MULSD,
+  ASM_MULSS,
   ASM_IMUL,
   ASM_MOV,
   ASM_MOVSX,
@@ -87,6 +89,8 @@ typedef enum {
   ASM_SHR,
   ASM_SUB,
   ASM_TEST,
+  ASM_UCOMISD,
+  ASM_UCOMISS,
   ASM_XCHG,
   ASM_XOR,
   // Registers, order matters.
@@ -159,6 +163,40 @@ typedef enum {
   ASM_R13,
   ASM_R14,
   ASM_R15,
+
+  ASM_MM0,
+  ASM_MM1,
+  ASM_MM2,
+  ASM_MM3,
+  ASM_MM4,
+  ASM_MM5,
+  ASM_MM6,
+  ASM_MM7,
+  ASM_MM_RESRVED_0, /* Reserved to preserve reg group size. */
+  ASM_MM_RESRVED_1,
+  ASM_MM_RESRVED_2,
+  ASM_MM_RESRVED_3,
+  ASM_MM_RESRVED_4,
+  ASM_MM_RESRVED_5,
+  ASM_MM_RESRVED_6,
+  ASM_MM_RESRVED_7,
+
+  ASM_XMM0,
+  ASM_XMM1,
+  ASM_XMM2,
+  ASM_XMM3,
+  ASM_XMM4,
+  ASM_XMM5,
+  ASM_XMM6,
+  ASM_XMM7,
+  ASM_XMM8,
+  ASM_XMM9,
+  ASM_XMM10,
+  ASM_XMM11,
+  ASM_XMM12,
+  ASM_XMM13,
+  ASM_XMM14,
+  ASM_XMM15,
   
   /* RIP is in a special class of its own. */
   ASM_RIP,
diff --git a/test/test.sh b/test/test.sh
index 9946f7f..8658e5f 100644
--- a/test/test.sh
+++ b/test/test.sh
@@ -30,6 +30,32 @@ t () {
   echo -n "."
 }
 
+t "mulsd %xmm0, %xmm1"
+t "mulsd (%rax), %xmm1"
+t "mulsd %xmm10, %xmm1"
+t "mulsd %xmm1, %xmm10"
+t "mulsd %xmm10, %xmm11"
+t "mulsd (%rax), %xmm11"
+t "mulss %xmm0, %xmm1"
+t "mulss (%rax), %xmm1"
+t "mulss %xmm10, %xmm1"
+t "mulss %xmm1, %xmm10"
+t "mulss %xmm10, %xmm11"
+t "mulss (%rax), %xmm11"
+
+
+t "ucomisd %xmm0, %xmm1"
+t "ucomisd (%rax), %xmm1"
+t "ucomisd %xmm10, %xmm1"
+t "ucomisd %xmm1, %xmm10"
+t "ucomisd %xmm10, %xmm11"
+t "ucomisd (%rax), %xmm11"
+t "ucomiss %xmm0, %xmm1"
+t "ucomiss (%rax), %xmm1"
+t "ucomiss %xmm10, %xmm1"
+t "ucomiss %xmm1, %xmm10"
+t "ucomiss %xmm10, %xmm11"
+t "ucomiss (%rax), %xmm11"
 
 
 for r in a b
@@ -57,7 +83,6 @@ do
   t "lea 2147483647(%r${r}x, %r${r}x, 4), %r${r}x"
 done
 
-
 t "ret"
 t "cltd"
 t "cqto"