fx.reflection: optimise linux dynamic function calling

fx.reflection/sys/linux/x86_64/callvm.S

@@ -1,3 +1,6 @@
+.extern memcpy
+.type memcpy, @function
+
 .global callvm_invoke_i
 .type callvm_invoke_i, @function
 
@@ -14,147 +17,62 @@ callvm_invoke_i:
 	push %r14
 	push %r15
 
+	# move our parameters out of the way
 	mov %rdi, %r11
 	mov %rsi, %r12
 
-	# First, set up the fixed arguments
-	# %r13 = fixed arg count
-	movq (%r12), %r13
-
-	# arg[0]
-	cmp $0, %r13
-	jle .idone
-	movq 48(%r12), %rdi
-
-	# arg[1]
-	cmp $1, %r13
-	jle .idone
-	movq 56(%r12), %rsi
-
-	# arg[2]
-	cmp $2, %r13
-	jle .idone
-	movq 64(%r12), %rdx
-
-	# arg[3]
-	cmp $3, %r13
-	jle .idone
-	movq 72(%r12), %rcx
-
-	# arg[4]
-	cmp $4, %r13
-	jle .idone
-	movq 80(%r12), %r8
-
-	# arg[5]
-	cmp $5, %r13
-	jle .idone
-	movq 88(%r12), %r9
-
-.idone:
-	# Next, set up the fixed double arguments
-	# r13 = fixed double arg count
-	movq 8(%r12), %r13
-
-	# arg[0]
-	cmp $0, %r13
-	jle .vdone
-	movq 96(%r12), %xmm0
-
-	# arg[1]
-	cmp $1, %r13
-	jle .vdone
-	movq 104(%r12), %xmm1
-
-	# arg[2]
-	cmp $2, %r13
-	jle .vdone
-	movq 112(%r12), %xmm2
-
-	# arg[3]
-	cmp $3, %r13
-	jle .vdone
-	movq 120(%r12), %xmm3
-
-	# arg[4]
-	cmp $4, %r13
-	jle .vdone
-	movq 128(%r12), %xmm4
-
-	# arg[5]
-	cmp $5, %r13
-	jle .vdone
-	movq 136(%r12), %xmm5
-
-	# arg[6]
-	cmp $6, %r13
-	jle .vdone
-	movq 144(%r12), %xmm6
-
-	# arg[7]
-	cmp $7, %r13
-	jle .vdone
-	movq 152(%r12), %xmm7
-
-.vdone:
-	# Finally, set up the variable arguments
-
-	# x8 = excess arg count
-	movq 32(%r12), %r13
-
 	# calculate the amount of stack space needed for the varargs
+	movq 32(%r12), %r13
 	shl $3, %r13
 
-	andq $0xFFFFFFFFFFFFFFF0, %r13
-	addq $0x10, %r13
-
 	# allocate the stack space
+	push %rsp
 	sub %r13, %rsp
-	mov %rsp, %r14
+	andq $0xFFFFFFFFFFFFFFF0, %rsp # re-align the stack
 
-	# convert buffer size back to number of arguments
-	movq 32(%r12), %r13
+	# copy the excess args to the stack
+	mov %rsp, %rdi
+	mov 160(%r12), %rsi
+	mov %r13, %rdx
+	call memcpy
 
-	# r13: number of var args (decrements with every iteration)
-	# r14: var arg dest pointer (increments with every iteration)
-	# r15: arg src buffer (increments with every iteration)
-	mov 160(%r12), %r15
+	# Next, set up the fixed integer arguments
 
-.loop:
-	cmp $0, %r13
-	je .loop_end
+	movq 48(%r12), %rdi	# int arg 0
+	movq 56(%r12), %rsi	# int arg 1
+	movq 64(%r12), %rdx	# int arg 2
+	movq 72(%r12), %rcx	# int arg 3
+	movq 80(%r12), %r8	# int arg 4
+	movq 88(%r12), %r9	# int arg 5
 
-	# read the arg value from the src pointer
-	mov (%r15), %rax
+	# Finally, set up the fixed double arguments
 
-	# write it to the stack, and increment the dest pointer
-	mov %rax, (%r14)
-	add $8, %r14
+	movq 96(%r12), %xmm0	# double arg 0
+	movq 104(%r12), %xmm1	# double arg 1
+	movq 112(%r12), %xmm2	# double arg 2
+	movq 120(%r12), %xmm3	# double arg 3
+	movq 128(%r12), %xmm4	# double arg 4
+	movq 136(%r12), %xmm5	# double arg 5
+	movq 144(%r12), %xmm6	# double arg 6
+	movq 152(%r12), %xmm7	# double arg 7
 
-	# increment the src pointer, decrement the arg count
-	sub $1, %r13
-	add $8, %r15
-
-	jmp .loop
-.loop_end:
+	# set the number of vararg double parameters
+	# as required by the ABI
+	mov 168(%r12), %rax
 
 	# call the function implementation
-	mov 168(%r12), %rax
 	call *%r11
 
-	# de-allocate the stack varargs buffer (the size is now stored in x19)
-	movq 32(%r12), %r13
-	shl $3, %r13
-	andq $0xFFFFFFFFFFFFFFF0, %r13
-	addq $0x10, %r13
-	addq %r13, %rsp
+	# Restore the stack pointer (deallocating the varargs buffer)
+	mov -40(%rbp), %rsp
 
+	# Restore callee-saved registers
 	pop %r15
 	pop %r14
 	pop %r13
 	pop %r12
 
-	# restore the saved stack frame and link pointer
+	# restore the saved stack frame
 	pop %rbp
 	ret