sketch vmNativeCall implementation for powerpc

src/powerpc.S

@@ -12,6 +12,7 @@
 
 .text
 
+BytesPerWord = 4
 LinkageArea = 24
 
 #ifdef __APPLE__
@@ -24,35 +25,118 @@ vmNativeCall:
    // save return address
    mflr r0
    stw  r0,8(r1)
-
+   
-   // r3 aka 24(r1): function
+   // r3 aka r13: function
-   // r4 aka 28(r1): stack
+   // r4        : stackTotal
-   // r5 aka 32(r1): stackSize
+   // r5        : memoryTable
-   // r6 aka 36(r1): gprTable
+   // r6        : memoryCount
-   // r7 aka 40(r1): fprTable
+   // r7        : gprTable
-   // r8 aka 48(r1): returnType
+   // r8        : fprTable
+   // r9 aka r14: returnType
+   
+   // r15       : stack frame size
+   // r16       : temporary
+   // r17       : temporary
+   
+   // save registers used for local variables
+   stw  r13,24(r1)
+   stw  r14,28(r1)
+   stw  r15,32(r1)
+   stw  r16,36(r1)
+   stw  r17,40(r1)
 
    // calculate aligned stack frame size
-   add  r13,r5,LinkageArea + 15
+   add  r15,r4,LinkageArea + LocalArea + 15
-   and  r13,r13,-16
+   and  r15,r15,-16
 
    // save and update stack pointer
-   sub  r14,r1,r9
+   sub  r16,r1,r15
-   stwu r1,r14
+   stwu r16,(r1)
 
    // save our argument registers so we can clobber them
-   stw  r3,24(r1)
+   mr   r13,r3
-   stw  r4,28(r1)
+   mr   r14,r9
-   stw  r5,32(r1)
-   stw  r6,36(r1)
-   stw  r7,40(r1)
-   stw  r8,48(r1)
-   
-   // todo
 
+   li   r16,0
+   b    test
+   
+loop:
+   lwzx r17,r16,r5
+   stwx r17,r16,r1
+   add  r16,r16,BytesPerWord
+   
+test:
+   cmp  r16,r6
+   blt  loop
+
+   // do we need to load the floating point registers?
+   cmp  r8,0
+   be   gpr
+
+   // yes, we do
+   lfd  fr1,0(r8)
+   lfd  fr2,8(r8)
+   lfd  fr3,16(r8)
+   lfd  fr4,24(r8)
+   lfd  fr5,32(r8)
+   lfd  fr6,40(r8)
+   lfd  fr7,48(r8)
+   lfd  fr8,56(r8)
+   lfd  fr9,64(r8)
+   lfd  fr10,72(r8)
+   lfd  fr11,80(r8)
+   lfd  fr12,88(r8)
+   lfd  fr13,96(r8)
+   
+gpr:
+   // do we need to load the general-purpose registers?
+   cmp  r7,0
+   be   call
+
+   // yes, we do
+   mr   r16,r7
+   lwz  r3,0(r16)
+   lwz  r4,4(r16)
+   lwz  r5,8(r16)
+   lwz  r6,12(r16)
+   lwz  r7,16(r16)
+   lwz  r8,20(r16)
+   lwz  r9,24(r16)
+   lwz  r10,28(r16)
+
+call:
+   // load and call function address
+   mtctr r13
+   bcctrl
+
+   // handle return value based on expected type
+   cmp  r14,VOID_TYPE
+   bne  float
+   b    exit
+
+float:
+   cmp  r14,FLOAT_TYPE
+   be   copy
+   cmp  r14,DOUBLE_TYPE
+   bne  exit
+
+copy:
+   // move floating point return value to GPRs via memory
+   sfd  fr1,8(r1)
+   lwz  r3,8(r1)
+   lwz  r4,12(r1)
+
+exit:   
    // restore stack pointer
    addi r1,r1,r13
 
+   // restore registers used for local variables
+   lwz  r13,24(r1)
+   lwz  r14,28(r1)
+   lwz  r15,32(r1)
+   lwz  r16,36(r1)
+   lwz  r17,40(r1)
+
    // load return address
    lwz  r0,8(r1)
    mtlr r0

src/powerpc.h

@@ -34,8 +34,9 @@
 #endif
 
 extern "C" uint64_t
-vmNativeCall(void* function, void* stack, unsigned stackSize,
+vmNativeCall(void* function, unsigned stackTotal, void* memoryTable,
-             void* gprTable, void* fprTable, unsigned returnType);
+             unsigned memoryCount, void* gprTable, void* fprTable,
+             unsigned returnType);
 
 namespace vm {
 
@@ -52,6 +53,7 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
   unsigned fprIndex = 0;
 
   uint64_t stack[argumentsSize];
+  unsigned stackSkip = 0;
   unsigned stackIndex = 0;
 
   for (unsigned i = 0; i < argumentsSize; ++i) {
@@ -59,7 +61,8 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
     case FLOAT_TYPE: {
       if (fprIndex < FprCount) {
         fprTable[fprIndex++] = arguments[i];
-        gprIndex += 1;
+        ++ gprIndex;
+        ++ stackSkip;
       } else {
         stack[stackIndex++] = arguments[i];
       }
@@ -68,8 +71,9 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
     case DOUBLE_TYPE: {
       if (fprIndex < FprCount) {
         memcpy(fprTable + fprIndex, arguments + i, 8);
-        fprIndex += BytesPerWord / 4;
+        ++ fprIndex;
         gprIndex += BytesPerWord / 4;
+        stackSkip += BytesPerWord / 4;
         i += (BytesPerWord / 4) - 1;
       } else {
         memcpy(stack + stackIndex, arguments + i, 8);
@@ -82,6 +86,7 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
       if (gprIndex < GprCount) {
         memcpy(gprTable + gprIndex, arguments + i, 8);
         gprIndex += BytesPerWord / 4;
+        stackSkip += BytesPerWord / 4;
         i += (BytesPerWord / 4) - 1;
       } else {
         memcpy(stack + stackIndex, arguments + i, 8);
@@ -93,6 +98,7 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
     default: {
       if (gprIndex < GprCount) {
         gprTable[gprIndex++] = arguments[i];
+        ++ stackSkip;
       } else {
         stack[stackIndex++] = arguments[i];
       }
@@ -100,7 +106,8 @@ dynamicCall(void* function, uintptr_t* arguments, uint8_t* argumentTypes,
     }
   }
 
-  return vmNativeCall(function, stack, stackIndex * BytesPerWord,
+  return vmNativeCall(function, (stackSkip + stackIndex) * BytesPerWord,
+                      stack, stackIndex,
                       (gprIndex ? gprTable : 0),
                       (fprIndex ? fprTable : 0), returnType);
 }