diff --git a/block_trampolines.S b/block_trampolines.S
index 9ebbeaf..cb2c78d 100644
--- a/block_trampolines.S
+++ b/block_trampolines.S
@@ -15,108 +15,97 @@
 # argument (the first block argument) and the _cmd parameter excised
 
 .file	"block_trampolines.S"
-#ifdef __ARM_ARCH
-#ifdef __arm__
-.syntax unified
-#endif
-.globl CDECL(__objc_block_trampoline_sret)
-TYPE_DIRECTIVE(CDECL(__objc_block_trampoline_sret), %function)
-.globl CDECL(__objc_block_trampoline_end_sret)
-.globl CDECL(__objc_block_trampoline)
-TYPE_DIRECTIVE(CDECL(__objc_block_trampoline), %function)
-.globl CDECL(__objc_block_trampoline_end)
-#else
-.globl CDECL(__objc_block_trampoline_sret)
-TYPE_DIRECTIVE(CDECL(__objc_block_trampoline_sret), @function)
-.globl CDECL(__objc_block_trampoline_end_sret)
-.globl CDECL(__objc_block_trampoline)
-TYPE_DIRECTIVE(CDECL(__objc_block_trampoline), @function)
-.globl CDECL(__objc_block_trampoline_end)
-#endif
+
+
 #if __x86_64
-#ifdef _WIN64
-#define ARG0 %rcx
-#define ARG1 %rdx
-#define ARG2 %r8
-#else
-#define ARG0 %rdi
-#define ARG1 %rsi
-#define ARG2 %rdx
-#endif
-CDECL(__objc_block_trampoline):
-	mov   -0x1007(%rip), ARG1      # Load the block pointer into the second argument
-	xchg  ARG1, ARG0           # Swap the first and second arguments
-	jmp   *-0x1008(%rip)           # Call the block function
-CDECL(__objc_block_trampoline_end):
-CDECL(__objc_block_trampoline_sret):
-	mov   -0x1007(%rip), ARG2      # Load the block pointer into the second argument
-	xchg  ARG1, ARG2           # Swap the first and second arguments
-	jmp   *-0x1008(%rip)           # Call the block function
-CDECL(__objc_block_trampoline_end_sret):
+////////////////////////////////////////////////////////////////////////////////
+// x86-64 trampoline
+////////////////////////////////////////////////////////////////////////////////
+.macro trampoline arg0, arg1
+	mov   -0x1007(%rip), \arg1   # Load the block pointer into the second argument
+	xchg  \arg1, \arg0           # Swap the first and second arguments
+	jmp   *-0x1008(%rip)         # Call the block function
+.endm
+// The Win64 and SysV x86-64 ABIs use different registers
+#	ifdef _WIN64
+#		define ARG0 %rcx
+#		define ARG1 %rdx
+#		define SARG1 %r8
+#	else
+#		define ARG0 %rdi
+#		define ARG1 %rsi
+#		define SARG1 %rdx
+#	endif
+#	define SARG0 ARG1
+
 #elif __i386
-CDECL(__objc_block_trampoline):
-	call  Lnext_line            # Store the instruction pointer on the stack
-Lnext_line:
-	pop   %eax                 # Load the old instruction pointer
-	mov   4(%esp), %ebx        # Load the self parameter
-	mov   %ebx, 8(%esp)        # Store self as the second argument
-	mov   -0x1005(%eax), %ebx       # Load the block pointer to %ebx
-	mov   %ebx, 4(%esp)        # Store the block pointer in the first argument
-	jmp   *-0x1001(%eax)           # Call the block function
-CDECL(__objc_block_trampoline_end):
-CDECL(__objc_block_trampoline_sret):
-	call  Lnext_line2           # Store the instruction pointer on the stack
-Lnext_line2:
+////////////////////////////////////////////////////////////////////////////////
+// x86-32 trampoline
+////////////////////////////////////////////////////////////////////////////////
+.macro trampoline arg0, arg1
+	call  1f                   # Store the instruction pointer on the stack
+1:
 	pop   %eax                 # Load the old instruction pointer
-	mov   8(%esp), %ebx        # Load the self parameter
-	mov   %ebx, 12(%esp)       # Store self as the second argument
-	mov   -0x1005(%eax), %ebx       # Load the block pointer to %ebx
-	mov   %ebx, 8(%esp)        # Store the block pointer in the first argument
-	jmp   *-0x1001(%eax)           # Call the block function
-CDECL(__objc_block_trampoline_end_sret):
+	mov   \arg0(%esp), %ebx    # Load the self parameter
+	mov   %ebx, \arg1(%esp)    # Store self as the second argument
+	mov   -0x1005(%eax), %ebx  # Load the block pointer to %ebx
+	mov   %ebx, \arg0(%esp)    # Store the block pointer in the first argument
+	jmp   *-0x1001(%eax)       # Call the block function
+.endm
+// All arguments on i386 are passed on the stack.  These values are stack
+// offsets - on other platforms they're register values.
+#	define ARG0 4
+#	define ARG1 8
+#	define SARG0 8
+#	define SARG1 12
+
 #elif __mips__
+////////////////////////////////////////////////////////////////////////////////
+// MIPS trampoline
+////////////////////////////////////////////////////////////////////////////////
 #	ifdef _ABI64
-CDECL(__objc_block_trampoline):
-	move     $a1, $a0
-	ld       $a0, -4096($25)
-	ld       $25, -4088($25)
-	jr       $25
-CDECL(__objc_block_trampoline_end):
-CDECL(__objc_block_trampoline_sret):
-	move     $a2, $a1
-	ld       $a1, -4096($25)
+.macro trampoline arg0, arg1
+	move     \arg1, \arg0
+	ld       \arg0, -4096($25)
 	ld       $25, -4088($25)
 	jr       $25
-CDECL(__objc_block_trampoline_end_sret):
+.endm
 #	else
-CDECL(__objc_block_trampoline):
-	move   $a1, $a0
-	lw     $a0, -4096($25)
-	lw     $25, -4092($25)
-	jr     $25
-CDECL(__objc_block_trampoline_end):
-CDECL(__objc_block_trampoline_sret):
-	move   $a2, $a1
-	lw     $a1, -4096($25)
+// 32-bit variant.  This ought to work with both n32 and o32, because they both
+// use 32-bit pointers and both use the same registers for the first four
+// arguments (and we only care about the first three).
+.macro trampoline arg0, arg1
+	move   \arg1, \arg0
+	lw     \arg0, -4096($25)
 	lw     $25, -4092($25)
 	jr     $25
-CDECL(__objc_block_trampoline_end_sret):
+.endm
 #	endif
+#define ARG0 $a0
+#define ARG1 $a1
+#define ARG2 $a2
+
 #elif defined(__ARM_ARCH_ISA_A64)
+////////////////////////////////////////////////////////////////////////////////
+// AArch64 (ARM64) trampoline
+////////////////////////////////////////////////////////////////////////////////
 .macro trampoline arg0, arg1
 	adr x17, #-4096
 	mov \arg1, \arg0
 	ldp \arg0, x17, [x17]
 	br x17
 .endm
-CDECL(__objc_block_trampoline):
-	trampoline x0, x1
-CDECL(__objc_block_trampoline_end):
-CDECL(__objc_block_trampoline_sret):
-	trampoline x1, x2
-CDECL(__objc_block_trampoline_end_sret):
+#define ARG0 x0
+#define ARG1 x1
+#define SARG0 x0
+#define SARG1 x1
+
 #elif __arm__
-#if 0 && ((__ARM_ARCH >= 7) || defined (__ARM_ARCH_6T2__))
+////////////////////////////////////////////////////////////////////////////////
+// AArch32 (ARM) trampoline
+////////////////////////////////////////////////////////////////////////////////
+
+#	if ((__ARM_ARCH >= 7) || defined (__ARM_ARCH_6T2__))
 // If we're on a target that supports Thumb 2, then we need slightly more
 // instructions to support Thumb/ARM code for the IMP and so we need to make
 // the trampolines thumb to be able to fit them in 16 bytes (they fit exactly
@@ -129,27 +118,44 @@ CDECL(__objc_block_trampoline_end_sret):
 	ldr r12, [r12, #-1]         // Jump to the block function
 	bx  r12
 .endm
-#else
+#	else
 .macro trampoline arg0, arg1
 	sub r12, pc, #4096
 	mov \arg0, \arg1            // Move self over _cmd
 	ldr \arg0, [r12, #-8]       // Load the block pointer over self
 	ldr pc, [r12, #-4]          // Jump to the block function
 .endm
-#endif // ((__ARM_ARCH >= 7) || defined (__ARM_ARCH_6T2__))
-CDECL(__objc_block_trampoline):
-	trampoline r0, r1
-CDECL(__objc_block_trampoline_end):
-CDECL(__objc_block_trampoline_sret):
-	trampoline r1, r2
-CDECL(__objc_block_trampoline_end_sret):
+#	endif // ((__ARM_ARCH >= 7) || defined (__ARM_ARCH_6T2__))
+#define ARG0 r0
+#define ARG1 r1
+#define SARG0 r1
+#define SARG1 r2
+
 #else
+
 #warning imp_implementationWithBlock() not implemented for your architecture
+.macro trampoline arg0, arg1
+.endm
+#define ARG0 0
+#define ARG1 0
+#define SARG0 0
+#define SARG1 0
+
+#endif
+
+
+.globl CDECL(__objc_block_trampoline)
 CDECL(__objc_block_trampoline):
+	trampoline ARG0, ARG1
+.globl CDECL(__objc_block_trampoline_end)
 CDECL(__objc_block_trampoline_end):
+.globl CDECL(__objc_block_trampoline_sret)
 CDECL(__objc_block_trampoline_sret):
+	trampoline SARG0, SARG1
+.globl CDECL(__objc_block_trampoline_end_sret)
 CDECL(__objc_block_trampoline_end_sret):
-#endif
+
+
 #ifdef __ELF__
 .section .note.GNU-stack,"",%progbits
 #endif