stk-code_catmod/lib/angelscript/source/as_callfunc_ppc.cpp
Sachith Hasaranga Seneviratne df67597ec7 added AS library
2014-03-16 14:44:05 +05:30

673 lines
26 KiB
C++

/*
AngelCode Scripting Library
Copyright (c) 2003-2013 Andreas Jonsson
This software is provided 'as-is', without any express or implied
warranty. In no event will the authors be held liable for any
damages arising from the use of this software.
Permission is granted to anyone to use this software for any
purpose, including commercial applications, and to alter it and
redistribute it freely, subject to the following restrictions:
1. The origin of this software must not be misrepresented; you
must not claim that you wrote the original software. If you use
this software in a product, an acknowledgment in the product
documentation would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and
must not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source
distribution.
The original version of this library can be located at:
http://www.angelcode.com/angelscript/
Andreas Jonsson
andreas@angelcode.com
*/
//
// as_callfunc_ppc.cpp
//
// These functions handle the actual calling of system functions
//
// This version is PPC specific
//
#include <stdio.h>
#include "as_config.h"
#ifndef AS_MAX_PORTABILITY
#ifdef AS_PPC
#include "as_callfunc.h"
#include "as_scriptengine.h"
#include "as_texts.h"
#include "as_tokendef.h"
#include "as_context.h"
#include <stdlib.h>
BEGIN_AS_NAMESPACE
// This part was originally written by Pecan Heber, June 2006, for
// use on MacOS X with 32bit PPC processor. He based the code on the
// code in as_callfunc_sh4.cpp
#define AS_PPC_MAX_ARGS 32
// The array used to send values to the correct places.
// Contains a byte of argTypes to indicate the register tYpe to load
// or zero if end of arguments
// The +1 is for when CallThis (object methods) is used
// Extra +1 when returning in memory
// Extra +1 in ppcArgsType to ensure zero end-of-args marker
// TODO: multithread: We need to remove these global variables for thread-safety
enum argTypes { ppcENDARG, ppcINTARG, ppcFLOATARG, ppcDOUBLEARG };
static asDWORD ppcArgs[2*AS_PPC_MAX_ARGS + 1 + 1];
// Using extern "C" because we use this symbol name in the assembly code
extern "C"
{
static asBYTE ppcArgsType[2*AS_PPC_MAX_ARGS + 1 + 1 + 1];
}
// NOTE: these values are for PowerPC 32 bit.
#define PPC_LINKAGE_SIZE (24) // how big the PPC linkage area is in a stack frame
#define PPC_NUM_REGSTORE (9) // how many registers of the PPC we need to store/restore for ppcFunc()
#define PPC_REGSTORE_SIZE (4*PPC_NUM_REGSTORE) // how many bytes are required for register store/restore
#define EXTRA_STACK_SIZE (PPC_LINKAGE_SIZE + PPC_REGSTORE_SIZE) // memory required, not including parameters, for the stack frame
#define PPC_STACK_SIZE(numParams) (-( ( ((((numParams)<8)?8:(numParams))<<2) + EXTRA_STACK_SIZE + 15 ) & ~15 )) // calculates the total stack size needed for ppcFunc64, must pad to 16bytes
// Loads all data into the correct places and calls the function.
// ppcArgsType is an array containing a byte type (enum argTypes) for each argument.
// stackArgSize is the size in bytes for how much data to put on the stack frame
extern "C" asQWORD ppcFunc(const asDWORD* argsPtr, int StackArgSize, asDWORD func);
asm(" .text\n"
" .align 2\n" // align the code to 1 << 2 = 4 bytes
" .globl _ppcFunc\n"
"_ppcFunc:\n"
// We're receiving the following parameters
// r3 : argsPtr
// r4 : StackArgSize
// r5 : func
// The following registers are used through out the function
// r31 : the address of the label address, as reference for all other labels
// r30 : temporary variable
// r29 : arg list pointer
// r28 : number of FPR registers used by the parameters
// r27 : the function pointer that will be called
// r26 : the location of the parameters for the call
// r25 : arg type list pointer
// r24 : temporary variable
// r23 : number of GPR registers used by the parameters
// r1 : this is stack pointer
// r0 : temporary variable
// f0 : temporary variable
// We need to store some of the registers for restoral before returning to caller
// lr - always stored in 8(r1) - this is the return address
// cr - not required to be stored, but if it is, its place is in 4(r1) - this is the condition register
// r1 - always stored in 0(r1) - this is the stack pointer
// r11
// r13 to r31
// f14 to f31
// Store register values and setup our stack frame
" mflr r0 \n" // move the return address into r0
" stw r0, 8(r1) \n" // Store the return address on the stack
" stmw r23, -36(r1) \n" // Store registers r23 to r31 on the stack
" stwux r1, r1, r4 \n" // Increase the stack with the needed space and store the original value in the destination
// Obtain an address that we'll use as our position of reference when obtaining addresses of other labels
" bl address \n"
"address: \n"
" mflr r31 \n"
// initial registers for the function
" mr r29, r3 \n" // (r29) args list
" mr r27, r5 \n" // load the function pointer to call. func actually holds the pointer to our function
" addi r26, r1, 24 \n" // setup the pointer to the parameter area to the function we're going to call
" sub r0, r0, r0 \n" // zero out r0
" mr r23, r0 \n" // zero out r23, which holds the number of used GPR registers
" mr r28, r0 \n" // zero our r22, which holds the number of used float registers
// load the global ppcArgsType which holds the types of arguments for each argument
" addis r25, r31, ha16(_ppcArgsType - address) \n" // load the upper 16 bits of the address to r25
" la r25, lo16(_ppcArgsType - address)(r25) \n" // load the lower 16 bits of the address to r25
" subi r25, r25, 1 \n" // since we increment r25 on its use, we'll pre-decrement it
// loop through the arguments
"ppcNextArg: \n"
" addi r25, r25, 1 \n" // increment r25, our arg type pointer
// switch based on the current argument type (0:end, 1:int, 2:float 3:double)
" lbz r24, 0(r25) \n" // load the current argument type (it's a byte)
" mulli r24, r24, 4 \n" // our jump table has 4 bytes per case (1 instruction)
" addis r30, r31, ha16(ppcTypeSwitch - address) \n" // load the address of the jump table for the switch
" la r30, lo16(ppcTypeSwitch - address)(r30) \n"
" add r0, r30, r24 \n" // offset by our argument type
" mtctr r0 \n" // load the jump address into CTR
" bctr \n" // jump into the jump table/switch
" nop \n"
// the jump table/switch based on the current argument type
"ppcTypeSwitch: \n"
" b ppcArgsEnd \n"
" b ppcArgIsInteger \n"
" b ppcArgIsFloat \n"
" b ppcArgIsDouble \n"
// when we get here we have finished processing all the arguments
// everything is ready to go to call the function
"ppcArgsEnd: \n"
" mtctr r27 \n" // the function pointer is stored in r27, load that into CTR
" bctrl \n" // call the function. We have to do it this way so that the LR gets the proper
" nop \n" // return value (the next instruction below). So we have to branch from CTR instead of LR.
// Restore registers and caller's stack frame, then return to caller
" lwz r1, 0(r1) \n" // restore the caller's stack pointer
" lwz r0, 8(r1) \n" // load in the caller's LR
" mtlr r0 \n" // restore the caller's LR
" lmw r23, -36(r1) \n" // restore registers r23 to r31 from the stack
" blr \n" // return back to the caller
" nop \n"
// Integer argument (GPR register)
"ppcArgIsInteger: \n"
" addis r30, r31, ha16(ppcLoadIntReg - address) \n" // load the address to the jump table for integer registers
" la r30, lo16(ppcLoadIntReg - address)(r30) \n"
" mulli r0, r23, 8 \n" // each item in the jump table is 2 instructions (8 bytes)
" add r0, r0, r30 \n" // calculate ppcLoadIntReg[numUsedGPRRegs]
" lwz r30, 0(r29) \n" // load the next argument from the argument list into r30
" cmpwi r23, 8 \n" // we can only load GPR3 through GPR10 (8 registers)
" bgt ppcLoadIntRegUpd \n" // if we're beyond 8 GPR registers, we're in the stack, go there
" mtctr r0 \n" // load the address of our ppcLoadIntReg jump table (we're below 8 GPR registers)
" bctr \n" // load the argument into a GPR register
" nop \n"
// jump table for GPR registers, for the first 8 GPR arguments
"ppcLoadIntReg: \n"
" mr r3, r30 \n" // arg0 (to r3)
" b ppcLoadIntRegUpd \n"
" mr r4, r30 \n" // arg1 (to r4)
" b ppcLoadIntRegUpd \n"
" mr r5, r30 \n" // arg2 (to r5)
" b ppcLoadIntRegUpd \n"
" mr r6, r30 \n" // arg3 (to r6)
" b ppcLoadIntRegUpd \n"
" mr r7, r30 \n" // arg4 (to r7)
" b ppcLoadIntRegUpd \n"
" mr r8, r30 \n" // arg5 (to r8)
" b ppcLoadIntRegUpd \n"
" mr r9, r30 \n" // arg6 (to r9)
" b ppcLoadIntRegUpd \n"
" mr r10, r30 \n" // arg7 (to r10)
" b ppcLoadIntRegUpd \n"
// all GPR arguments still go on the stack
"ppcLoadIntRegUpd: \n"
" stw r30, 0(r26) \n" // store the argument into the next slot on the stack's argument list
" addi r23, r23, 1 \n" // count a used GPR register
" addi r29, r29, 4 \n" // move to the next argument on the list
" addi r26, r26, 4 \n" // adjust our argument stack pointer for the next
" b ppcNextArg \n" // next argument
// single Float argument
"ppcArgIsFloat:\n"
" addis r30, r31, ha16(ppcLoadFloatReg - address) \n" // get the base address of the float register jump table
" la r30, lo16(ppcLoadFloatReg - address)(r30) \n"
" mulli r0, r28, 8 \n" // each jump table entry is 8 bytes
" add r0, r0, r30 \n" // calculate the offset to ppcLoadFloatReg[numUsedFloatReg]
" lfs f0, 0(r29) \n" // load the next argument as a float into f0
" cmpwi r28, 13 \n" // can't load more than 13 float/double registers
" bgt ppcLoadFloatRegUpd \n" // if we're beyond 13 registers, just fall to inserting into the stack
" mtctr r0 \n" // jump into the float jump table
" bctr \n"
" nop \n"
// jump table for float registers, for the first 13 float arguments
"ppcLoadFloatReg: \n"
" fmr f1, f0 \n" // arg0 (f1)
" b ppcLoadFloatRegUpd \n"
" fmr f2, f0 \n" // arg1 (f2)
" b ppcLoadFloatRegUpd \n"
" fmr f3, f0 \n" // arg2 (f3)
" b ppcLoadFloatRegUpd \n"
" fmr f4, f0 \n" // arg3 (f4)
" b ppcLoadFloatRegUpd \n"
" fmr f5, f0 \n" // arg4 (f5)
" b ppcLoadFloatRegUpd \n"
" fmr f6, f0 \n" // arg5 (f6)
" b ppcLoadFloatRegUpd \n"
" fmr f7, f0 \n" // arg6 (f7)
" b ppcLoadFloatRegUpd \n"
" fmr f8, f0 \n" // arg7 (f8)
" b ppcLoadFloatRegUpd \n"
" fmr f9, f0 \n" // arg8 (f9)
" b ppcLoadFloatRegUpd \n"
" fmr f10, f0 \n" // arg9 (f10)
" b ppcLoadFloatRegUpd \n"
" fmr f11, f0 \n" // arg10 (f11)
" b ppcLoadFloatRegUpd \n"
" fmr f12, f0 \n" // arg11 (f12)
" b ppcLoadFloatRegUpd \n"
" fmr f13, f0 \n" // arg12 (f13)
" b ppcLoadFloatRegUpd \n"
" nop \n"
// all float arguments still go on the stack
"ppcLoadFloatRegUpd: \n"
" stfs f0, 0(r26) \n" // store, as a single float, f0 (current argument) on to the stack argument list
" addi r23, r23, 1 \n" // a float register eats up a GPR register
" addi r28, r28, 1 \n" // ...and, of course, a float register
" addi r29, r29, 4 \n" // move to the next argument in the list
" addi r26, r26, 4 \n" // move to the next stack slot
" b ppcNextArg \n" // on to the next argument
" nop \n"
// double Float argument
"ppcArgIsDouble: \n"
" addis r30, r31, ha16(ppcLoadDoubleReg - address) \n" // load the base address of the jump table for double registers
" la r30, lo16(ppcLoadDoubleReg - address)(r30) \n"
" mulli r0, r28, 8 \n" // each slot of the jump table is 8 bytes
" add r0, r0, r30 \n" // calculate ppcLoadDoubleReg[numUsedFloatReg]
" lfd f0, 0(r29) \n" // load the next argument, as a double float, into f0
" cmpwi r28, 13 \n" // the first 13 floats must go into float registers also
" bgt ppcLoadDoubleRegUpd \n" // if we're beyond 13, then just put on to the stack
" mtctr r0 \n" // we're under 13, first load our register
" bctr \n" // jump into the jump table
" nop \n"
// jump table for float registers, for the first 13 float arguments
"ppcLoadDoubleReg: \n"
" fmr f1, f0 \n" // arg0 (f1)
" b ppcLoadDoubleRegUpd \n"
" fmr f2, f0 \n" // arg1 (f2)
" b ppcLoadDoubleRegUpd \n"
" fmr f3, f0 \n" // arg2 (f3)
" b ppcLoadDoubleRegUpd \n"
" fmr f4, f0 \n" // arg3 (f4)
" b ppcLoadDoubleRegUpd \n"
" fmr f5, f0 \n" // arg4 (f5)
" b ppcLoadDoubleRegUpd \n"
" fmr f6, f0 \n" // arg5 (f6)
" b ppcLoadDoubleRegUpd \n"
" fmr f7, f0 \n" // arg6 (f7)
" b ppcLoadDoubleRegUpd \n"
" fmr f8, f0 \n" // arg7 (f8)
" b ppcLoadDoubleRegUpd \n"
" fmr f9, f0 \n" // arg8 (f9)
" b ppcLoadDoubleRegUpd \n"
" fmr f10, f0 \n" // arg9 (f10)
" b ppcLoadDoubleRegUpd \n"
" fmr f11, f0 \n" // arg10 (f11)
" b ppcLoadDoubleRegUpd \n"
" fmr f12, f0 \n" // arg11 (f12)
" b ppcLoadDoubleRegUpd \n"
" fmr f13, f0 \n" // arg12 (f13)
" b ppcLoadDoubleRegUpd \n"
" nop \n"
// all float arguments still go on the stack
"ppcLoadDoubleRegUpd: \n"
" stfd f0, 0(r26) \n" // store f0, as a double, into the argument list on the stack
" addi r23, r23, 2 \n" // a double float eats up two GPRs
" addi r28, r28, 1 \n" // ...and, of course, a float
" addi r29, r29, 8 \n" // increment to our next argument we need to process (8 bytes for the 64bit float)
" addi r26, r26, 8 \n" // increment to the next slot on the argument list on the stack (8 bytes)
" b ppcNextArg \n" // on to the next argument
" nop \n"
);
asDWORD GetReturnedFloat()
{
asDWORD f;
asm(" stfs f1, %0\n" : "=m"(f));
return f;
}
asQWORD GetReturnedDouble()
{
asQWORD f;
asm(" stfd f1, %0\n" : "=m"(f));
return f;
}
// puts the arguments in the correct place in the stack array. See comments above.
void stackArgs(const asDWORD *args, const asBYTE *argsType, int& numIntArgs, int& numFloatArgs, int& numDoubleArgs)
{
int i;
int argWordPos = numIntArgs + numFloatArgs + (numDoubleArgs*2);
int typeOffset = numIntArgs + numFloatArgs + numDoubleArgs;
int typeIndex;
for( i = 0, typeIndex = 0; ; i++, typeIndex++ )
{
// store the type
ppcArgsType[typeOffset++] = argsType[typeIndex];
if( argsType[typeIndex] == ppcENDARG )
break;
switch( argsType[typeIndex] )
{
case ppcFLOATARG:
// stow float
ppcArgs[argWordPos] = args[i]; // it's just a bit copy
numFloatArgs++;
argWordPos++; //add one word
break;
case ppcDOUBLEARG:
// stow double
memcpy( &ppcArgs[argWordPos], &args[i], sizeof(double) ); // we have to do this because of alignment
numDoubleArgs++;
argWordPos+=2; //add two words
i++;//doubles take up 2 argument slots
break;
case ppcINTARG:
// stow register
ppcArgs[argWordPos] = args[i];
numIntArgs++;
argWordPos++;
break;
}
}
// close off the argument list (if we have max args we won't close it off until here)
ppcArgsType[typeOffset] = ppcENDARG;
}
static asQWORD CallCDeclFunction(const asDWORD* pArgs, const asBYTE *pArgsType, int argSize, asDWORD func, void *retInMemory)
{
int baseArgCount = 0;
if( retInMemory )
{
// the first argument is the 'return in memory' pointer
ppcArgs[0] = (asDWORD)retInMemory;
ppcArgsType[0] = ppcINTARG;
ppcArgsType[1] = ppcENDARG;
baseArgCount = 1;
}
// put the arguments in the correct places in the ppcArgs array
int numTotalArgs = baseArgCount;
if( argSize > 0 )
{
int intArgs = baseArgCount, floatArgs = 0, doubleArgs = 0;
stackArgs( pArgs, pArgsType, intArgs, floatArgs, doubleArgs );
numTotalArgs = intArgs + floatArgs + 2*doubleArgs; // doubles occupy two slots
}
else
{
// no arguments, cap the type list
ppcArgsType[baseArgCount] = ppcENDARG;
}
// call the function with the arguments
return ppcFunc( ppcArgs, PPC_STACK_SIZE(numTotalArgs), func );
}
// This function is identical to CallCDeclFunction, with the only difference that
// the value in the first parameter is the object (unless we are returning in memory)
static asQWORD CallThisCallFunction(const void *obj, const asDWORD* pArgs, const asBYTE *pArgsType, int argSize, asDWORD func, void *retInMemory )
{
int baseArgCount = 0;
if( retInMemory )
{
// the first argument is the 'return in memory' pointer
ppcArgs[0] = (asDWORD)retInMemory;
ppcArgsType[0] = ppcINTARG;
ppcArgsType[1] = ppcENDARG;
baseArgCount = 1;
}
// the first argument is the 'this' of the object
ppcArgs[baseArgCount] = (asDWORD)obj;
ppcArgsType[baseArgCount++] = ppcINTARG;
ppcArgsType[baseArgCount] = ppcENDARG;
// put the arguments in the correct places in the ppcArgs array
int numTotalArgs = baseArgCount;
if( argSize > 0 )
{
int intArgs = baseArgCount, floatArgs = 0, doubleArgs = 0;
stackArgs( pArgs, pArgsType, intArgs, floatArgs, doubleArgs );
numTotalArgs = intArgs + floatArgs + 2*doubleArgs; // doubles occupy two slots
}
// call the function with the arguments
return ppcFunc( ppcArgs, PPC_STACK_SIZE(numTotalArgs), func);
}
// This function is identical to CallCDeclFunction, with the only difference that
// the value in the last parameter is the object
// NOTE: on PPC the order for the args is reversed
static asQWORD CallThisCallFunction_objLast(const void *obj, const asDWORD* pArgs, const asBYTE *pArgsType, int argSize, asDWORD func, void *retInMemory)
{
UNUSED_VAR(argSize);
int baseArgCount = 0;
if( retInMemory )
{
// the first argument is the 'return in memory' pointer
ppcArgs[0] = (asDWORD)retInMemory;
ppcArgsType[0] = ppcINTARG;
ppcArgsType[1] = ppcENDARG;
baseArgCount = 1;
}
// stack any of the arguments
int intArgs = baseArgCount, floatArgs = 0, doubleArgs = 0;
stackArgs( pArgs, pArgsType, intArgs, floatArgs, doubleArgs );
int numTotalArgs = intArgs + floatArgs + doubleArgs;
// can we fit the object in at the end?
if( numTotalArgs < AS_PPC_MAX_ARGS )
{
// put the object pointer at the end
int argPos = intArgs + floatArgs + (doubleArgs * 2);
ppcArgs[argPos] = (asDWORD)obj;
ppcArgsType[numTotalArgs++] = ppcINTARG;
ppcArgsType[numTotalArgs] = ppcENDARG;
}
// call the function with the arguments
return ppcFunc( ppcArgs, PPC_STACK_SIZE(numTotalArgs), func );
}
asQWORD CallSystemFunctionNative(asCContext *context, asCScriptFunction *descr, void *obj, asDWORD *args, void *retPointer, asQWORD &/*retQW2*/)
{
// use a working array of types, we'll configure the final one in stackArgs
asBYTE argsType[2*AS_PPC_MAX_ARGS + 1 + 1 + 1];
memset( argsType, 0, sizeof(argsType));
asCScriptEngine *engine = context->m_engine;
asSSystemFunctionInterface *sysFunc = descr->sysFuncIntf;
asQWORD retQW = 0;
void *func = (void*)sysFunc->func;
int paramSize = sysFunc->paramSize;
asDWORD *vftable = NULL;
int a, s;
// convert the parameters that are < 4 bytes from little endian to big endian
int argDwordOffset = 0;
for( a = 0; a < (int)descr->parameterTypes.GetLength(); a++ )
{
int numBytes = descr->parameterTypes[a].GetSizeInMemoryBytes();
if( numBytes >= 4 || descr->parameterTypes[a].IsReference() || descr->parameterTypes[a].IsObjectHandle() )
{
argDwordOffset += descr->parameterTypes[a].GetSizeOnStackDWords();
continue;
}
// flip
asASSERT( numBytes == 1 || numBytes == 2 );
switch( numBytes )
{
case 1:
{
volatile asBYTE *bPtr = (asBYTE*)ARG_DW(args[argDwordOffset]);
asBYTE t = bPtr[0];
bPtr[0] = bPtr[3];
bPtr[3] = t;
t = bPtr[1];
bPtr[1] = bPtr[2];
bPtr[2] = t;
}
break;
case 2:
{
volatile asWORD *wPtr = (asWORD*)ARG_DW(args[argDwordOffset]);
asWORD t = wPtr[0];
wPtr[0] = wPtr[1];
wPtr[1] = t;
}
break;
}
argDwordOffset++;
}
// mark all float/double/int arguments
if( !sysFunc->takesObjByVal )
{
for( s = 0, a = 0; s < (int)descr->parameterTypes.GetLength(); s++, a++ )
{
if( descr->parameterTypes[s].IsFloatType() && !descr->parameterTypes[s].IsReference() )
{
argsType[a] = ppcFLOATARG;
}
else if( descr->parameterTypes[s].IsDoubleType() && !descr->parameterTypes[s].IsReference() )
{
argsType[a] = ppcDOUBLEARG;
}
else
{
argsType[a] = ppcINTARG;
if( descr->parameterTypes[s].GetSizeOnStackDWords() == 2 )
{
// Add an extra integer argument for the extra size
a++;
argsType[a] = ppcINTARG;
}
}
}
}
asDWORD paramBuffer[64];
if( sysFunc->takesObjByVal )
{
paramSize = 0;
int spos = 0;
int dpos = 1;
int a = 0;
for( asUINT n = 0; n < descr->parameterTypes.GetLength(); n++ )
{
if( descr->parameterTypes[n].IsObject() && !descr->parameterTypes[n].IsObjectHandle() && !descr->parameterTypes[n].IsReference() )
{
#ifdef COMPLEX_OBJS_PASSED_BY_REF
if( descr->parameterTypes[n].GetObjectType()->flags & COMPLEX_MASK )
{
argsType[a++] = ppcINTARG;
paramBuffer[dpos++] = args[spos++];
paramSize++;
}
else
#endif
{
// TODO: Probably have to handle asOBJ_APP_FLOAT as a primitive
// Copy the object's memory to the buffer
memcpy( &paramBuffer[dpos], *(void**)(args+spos), descr->parameterTypes[n].GetSizeInMemoryBytes() );
// Delete the original memory
engine->CallFree(*(char**)(args+spos) );
spos++;
asUINT dwords = descr->parameterTypes[n].GetSizeInMemoryDWords();
dpos += dwords;
paramSize += dwords;
for( asUINT i = 0; i < dwords; i++ )
argsType[a++] = ppcINTARG;
}
}
else
{
// Copy the value directly
paramBuffer[dpos++] = args[spos++];
if( descr->parameterTypes[n].IsFloatType() && !descr->parameterTypes[n].IsReference() )
argsType[a++] = ppcFLOATARG;
else if( descr->parameterTypes[n].IsDoubleType() && !descr->parameterTypes[n].IsReference() )
argsType[a++] = ppcDOUBLEARG;
else
argsType[a++] = ppcINTARG;
if( descr->parameterTypes[n].GetSizeOnStackDWords() > 1 )
{
paramBuffer[dpos++] = args[spos++];
if( !descr->parameterTypes[n].IsDoubleType() ) // Double already knows it is 2 dwords
argsType[a++] = ppcINTARG;
}
paramSize += descr->parameterTypes[n].GetSizeOnStackDWords();
}
}
// Keep a free location at the beginning
args = &paramBuffer[1];
}
int callConv = sysFunc->callConv;
switch( callConv )
{
case ICC_CDECL:
case ICC_CDECL_RETURNINMEM:
case ICC_STDCALL:
case ICC_STDCALL_RETURNINMEM:
retQW = CallCDeclFunction( args, argsType, paramSize, (asDWORD)func, retPointer );
break;
case ICC_THISCALL:
case ICC_THISCALL_RETURNINMEM:
retQW = CallThisCallFunction(obj, args, argsType, paramSize, (asDWORD)func, retPointer );
break;
case ICC_VIRTUAL_THISCALL:
case ICC_VIRTUAL_THISCALL_RETURNINMEM:
// Get virtual function table from the object pointer
vftable = *(asDWORD**)obj;
retQW = CallThisCallFunction( obj, args, argsType, paramSize, vftable[asDWORD(func)>>2], retPointer );
break;
case ICC_CDECL_OBJLAST:
case ICC_CDECL_OBJLAST_RETURNINMEM:
retQW = CallThisCallFunction_objLast( obj, args, argsType, paramSize, (asDWORD)func, retPointer );
break;
case ICC_CDECL_OBJFIRST:
case ICC_CDECL_OBJFIRST_RETURNINMEM:
retQW = CallThisCallFunction( obj, args, argsType, paramSize, (asDWORD)func, retPointer );
break;
default:
context->SetInternalException(TXT_INVALID_CALLING_CONVENTION);
}
// If the return is a float value we need to get the value from the FP register
if( sysFunc->hostReturnFloat )
{
if( sysFunc->hostReturnSize == 1 )
*(asDWORD*)&retQW = GetReturnedFloat();
else
retQW = GetReturnedDouble();
}
return retQW;
}
END_AS_NAMESPACE
#endif // AS_PPC
#endif // AS_MAX_PORTABILITY