NS_InvokeByIndex(nsISupports* that, PRUint32 methodIndex,
PRUint32 paramCount, nsXPTCVariant* params)
{
vtable_func *vtable = *reinterpret_cast<vtable_func **>(that);
vtable_func method = vtable[methodIndex];
PRUint32 overflow = invoke_count_words (paramCount, params);
PRUint32 *stack_space = reinterpret_cast<PRUint32 *>(__builtin_alloca((overflow + 8 /* 4 32-bits gpr + 2 64-bits fpr */) * 4));
invoke_copy_to_stack(paramCount, params, stack_space, overflow);
PRUint32 *d_gpr = stack_space + overflow;
double *d_fpr = reinterpret_cast<double *>(d_gpr + 4);
return method(that, d_gpr[0], d_gpr[1], d_gpr[2], d_gpr[3], d_fpr[0], d_fpr[1]);
}
NS_InvokeByIndex(nsISupports* that, uint32_t methodIndex,
uint32_t paramCount, nsXPTCVariant* params)
{
vtable_func *vtable = *reinterpret_cast<vtable_func **>(that);
vtable_func method = vtable[methodIndex];
uint64_t overflow = invoke_count_words (paramCount, params);
uint64_t *stack_space = reinterpret_cast<uint64_t *>(__builtin_alloca((overflow + 8 /* 4 64-bits gpr + 4 64-bits fpr */) * 8));
uint64_t result;
invoke_copy_to_stack(paramCount, params, stack_space, overflow);
uint64_t *d_gpr = stack_space + overflow;
double *d_fpr = reinterpret_cast<double *>(d_gpr + 4);
return method(that, d_gpr[0], d_gpr[1], d_gpr[2], d_gpr[3], d_fpr[0], d_fpr[1], d_fpr[2], d_fpr[3]);
}
NS_InvokeByIndex_P(nsISupports * that, uint32_t methodIndex,
uint32_t paramCount, nsXPTCVariant * params)
{
uint32_t nr_gpr, nr_fpr, nr_stack;
invoke_count_words(paramCount, params, nr_gpr, nr_fpr, nr_stack);
// Stack, if used, must be 16-bytes aligned
if (nr_stack)
nr_stack = (nr_stack + 1) & ~1;
// Load parameters to stack, if necessary
uint64_t *stack = (uint64_t *) __builtin_alloca(nr_stack * 8);
uint64_t gpregs[GPR_COUNT];
double fpregs[FPR_COUNT];
invoke_copy_to_stack(stack, paramCount, params, gpregs, fpregs);
// Load FPR registers from fpregs[]
register double d0 asm("xmm0");
register double d1 asm("xmm1");
register double d2 asm("xmm2");
register double d3 asm("xmm3");
register double d4 asm("xmm4");
register double d5 asm("xmm5");
register double d6 asm("xmm6");
register double d7 asm("xmm7");
switch (nr_fpr) {
#define ARG_FPR(N) \
case N+1: d##N = fpregs[N];
ARG_FPR(7);
ARG_FPR(6);
ARG_FPR(5);
ARG_FPR(4);
ARG_FPR(3);
ARG_FPR(2);
ARG_FPR(1);
ARG_FPR(0);
case 0:;
#undef ARG_FPR
}
// Load GPR registers from gpregs[]
register uint64_t a0 asm("rdi");
register uint64_t a1 asm("rsi");
register uint64_t a2 asm("rdx");
register uint64_t a3 asm("rcx");
register uint64_t a4 asm("r8");
register uint64_t a5 asm("r9");
switch (nr_gpr) {
#define ARG_GPR(N) \
case N+1: a##N = gpregs[N];
ARG_GPR(5);
ARG_GPR(4);
ARG_GPR(3);
ARG_GPR(2);
ARG_GPR(1);
case 1: a0 = (uint64_t) that;
case 0:;
#undef ARG_GPR
}
// Ensure that assignments to SSE registers won't be optimized away
asm("" ::
"x" (d0), "x" (d1), "x" (d2), "x" (d3),
"x" (d4), "x" (d5), "x" (d6), "x" (d7));
// Get pointer to method
uint64_t methodAddress = *((uint64_t *)that);
methodAddress += 8 * methodIndex;
methodAddress = *((uint64_t *)methodAddress);
typedef uint32_t (*Method)(uint64_t, uint64_t, uint64_t, uint64_t, uint64_t, uint64_t);
uint32_t result = ((Method)methodAddress)(a0, a1, a2, a3, a4, a5);
return result;
}
NS_InvokeByIndex(nsISupports * that, uint32_t methodIndex,
uint32_t paramCount, nsXPTCVariant * params)
{
uint32_t nr_stack;
invoke_count_words(paramCount, params, nr_stack);
// Stack, if used, must be 16-bytes aligned
if (nr_stack)
nr_stack = (nr_stack + 1) & ~1;
// Load parameters to stack, if necessary
uint64_t *stack = (uint64_t *) __builtin_alloca(nr_stack * 8);
uint64_t gpregs[GPR_COUNT];
double fpregs[FPR_COUNT];
invoke_copy_to_stack(stack, paramCount, params, gpregs, fpregs);
// We used to have switches to make sure we would only load the registers
// that are needed for this call. That produced larger code that was
// not faster in practice. It also caused compiler warnings about the
// variables being used uninitialized.
// We now just load every every register. There could still be a warning
// from a memory analysis tools that we are loading uninitialized stack
// positions.
// FIXME: this function depends on the above __builtin_alloca placing
// the array in the correct spot for the ABI.
// Load FPR registers from fpregs[]
double d0, d1, d2, d3, d4, d5, d6, d7;
d7 = fpregs[7];
d6 = fpregs[6];
d5 = fpregs[5];
d4 = fpregs[4];
d3 = fpregs[3];
d2 = fpregs[2];
d1 = fpregs[1];
d0 = fpregs[0];
// Load GPR registers from gpregs[]
uint64_t a0, a1, a2, a3, a4, a5;
a5 = gpregs[5];
a4 = gpregs[4];
a3 = gpregs[3];
a2 = gpregs[2];
a1 = gpregs[1];
a0 = (uint64_t) that;
// Get pointer to method
uint64_t methodAddress = *((uint64_t *)that);
methodAddress += 8 * methodIndex;
methodAddress = *((uint64_t *)methodAddress);
typedef nsresult (*Method)(uint64_t, uint64_t, uint64_t, uint64_t,
uint64_t, uint64_t, double, double, double,
double, double, double, double, double);
nsresult result = ((Method)methodAddress)(a0, a1, a2, a3, a4, a5,
d0, d1, d2, d3, d4, d5,
d6, d7);
return result;
}
NS_InvokeByIndex_P(nsISupports * that, PRUint32 methodIndex,
PRUint32 paramCount, nsXPTCVariant * params)
{
PRUint32 nr_gpr, nr_fpr, nr_stack;
invoke_count_words(paramCount, params, nr_gpr, nr_fpr, nr_stack);
// Stack, if used, must be 16-bytes aligned
if (nr_stack)
nr_stack = (nr_stack + 1) & ~1;
// Load parameters to stack, if necessary
PRUint64 *stack = (PRUint64 *) __builtin_alloca(nr_stack * 8);
PRUint64 gpregs[GPR_COUNT];
double fpregs[FPR_COUNT];
invoke_copy_to_stack(stack, paramCount, params, gpregs, fpregs);
// Load FPR registers from fpregs[]
double d0, d1, d2, d3, d4, d5, d6, d7;
switch (nr_fpr) {
#define ARG_FPR(N) \
case N+1: d##N = fpregs[N];
ARG_FPR(7);
ARG_FPR(6);
ARG_FPR(5);
ARG_FPR(4);
ARG_FPR(3);
ARG_FPR(2);
ARG_FPR(1);
ARG_FPR(0);
case 0:;
#undef ARG_FPR
}
// Load GPR registers from gpregs[]
PRUint64 a0, a1, a2, a3, a4, a5;
switch (nr_gpr) {
#define ARG_GPR(N) \
case N+1: a##N = gpregs[N];
ARG_GPR(5);
ARG_GPR(4);
ARG_GPR(3);
ARG_GPR(2);
ARG_GPR(1);
case 1: a0 = (PRUint64) that;
case 0:;
#undef ARG_GPR
}
// Get pointer to method
PRUint64 methodAddress = *((PRUint64 *)that);
methodAddress += 8 * methodIndex;
methodAddress = *((PRUint64 *)methodAddress);
typedef PRUint32 (*Method)(PRUint64, PRUint64, PRUint64, PRUint64,
PRUint64, PRUint64, double, double, double,
double, double, double, double, double);
PRUint32 result = ((Method)methodAddress)(a0, a1, a2, a3, a4, a5,
d0, d1, d2, d3, d4, d5,
d6, d7);
return result;
}
XPTC_InvokeByIndex(nsISupports * that, PRUint32 methodIndex,
PRUint32 paramCount, nsXPTCVariant * params)
{
PRUint32 nr_gpr, nr_fpr, nr_stack;
invoke_count_words(paramCount, params, nr_gpr, nr_fpr, nr_stack);
// Stack, if used, must be 16-bytes aligned
if (nr_stack)
nr_stack = (nr_stack + 1) & ~1;
// Load parameters to stack, if necessary
PRUint64 *stack = (PRUint64 *) __builtin_alloca(nr_stack * 8);
PRUint64 gpregs[GPR_COUNT];
double fpregs[FPR_COUNT];
invoke_copy_to_stack(stack, paramCount, params, gpregs, fpregs);
// disable the warning about sometimes not initialized variables which is hit
// when we pass less than 8 XMM or less than 6 GPR registers.
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wsometimes-uninitialized"
// Load FPR registers from fpregs[]
register double d0 asm("xmm0");
register double d1 asm("xmm1");
register double d2 asm("xmm2");
register double d3 asm("xmm3");
register double d4 asm("xmm4");
register double d5 asm("xmm5");
register double d6 asm("xmm6");
register double d7 asm("xmm7");
switch (nr_fpr) {
#define ARG_FPR(N) \
case N+1: d##N = fpregs[N];
ARG_FPR(7);
ARG_FPR(6);
ARG_FPR(5);
ARG_FPR(4);
ARG_FPR(3);
ARG_FPR(2);
ARG_FPR(1);
ARG_FPR(0);
case 0:;
#undef ARG_FPR
}
// Load GPR registers from gpregs[]
register PRUint64 a0 asm("rdi");
register PRUint64 a1 asm("rsi");
register PRUint64 a2 asm("rdx");
register PRUint64 a3 asm("rcx");
register PRUint64 a4 asm("r8");
register PRUint64 a5 asm("r9");
switch (nr_gpr) {
#define ARG_GPR(N) \
case N+1: a##N = gpregs[N];
ARG_GPR(5);
ARG_GPR(4);
ARG_GPR(3);
ARG_GPR(2);
ARG_GPR(1);
case 1: a0 = (PRUint64) that;
case 0:;
#undef ARG_GPR
}
// Ensure that assignments to SSE registers won't be optimized away
asm("" ::
"x" (d0), "x" (d1), "x" (d2), "x" (d3),
"x" (d4), "x" (d5), "x" (d6), "x" (d7));
// Get pointer to method
PRUint64 methodAddress = *((PRUint64 *)that);
methodAddress += 8 * methodIndex;
methodAddress = *((PRUint64 *)methodAddress);
typedef PRUint32 (*Method)(PRUint64, PRUint64, PRUint64, PRUint64, PRUint64, PRUint64);
PRUint32 result = ((Method)methodAddress)(a0, a1, a2, a3, a4, a5);
return result;
#pragma clang diagnostic pop
}