auto merge of #5418 : luqmana/rust/stack-float, r=brson

Like I commented in #2043, I can't reproduce the weirdness from #1388 on either mac or linux (x84_64) and pushing to try gives all green.

That's 128 less bytes to have to keep in the stack for every call to __morestack.

src/rt/arch/x86_64/morestack.S

@@ -54,7 +54,7 @@ MORESTACK:
 	// Calculate the CFA as on offset from %ebp
 	.cfi_def_cfa_register %rbp
 
-        subq $184, %rsp
+        subq $56, %rsp
 
 	// Save argument registers of the original function
 	movq %rdi,       (%rsp)
@@ -63,14 +63,6 @@ MORESTACK:
 	movq %rcx,     24(%rsp)
 	movq %r8,      32(%rsp)
 	movq %r9,      40(%rsp)
-	movdqa %xmm0,  48(%rsp)
-	movdqa %xmm1,  64(%rsp)
-	movdqa %xmm2,  80(%rsp)
-	movdqa %xmm3,  96(%rsp)
-	movdqa %xmm4, 112(%rsp)
-	movdqa %xmm5, 128(%rsp)
-	movdqa %xmm6, 144(%rsp)
-	movdqa %xmm7, 160(%rsp)
 
 	// Calculate the address of the stack arguments.
 	// We have the base pointer, __morestack's return address,
@@ -101,16 +93,8 @@ MORESTACK:
 	movq    24(%rsp), %rcx
 	movq    32(%rsp), %r8
 	movq    40(%rsp), %r9
-	movdqa  48(%rsp), %xmm0
-	movdqa  64(%rsp), %xmm1
-	movdqa  80(%rsp), %xmm2
-	movdqa  96(%rsp), %xmm3
-	movdqa 112(%rsp), %xmm4
-	movdqa 128(%rsp), %xmm5
-	movdqa 144(%rsp), %xmm6
-	movdqa 160(%rsp), %xmm7
-
-	addq $184, %rsp
+
+	addq $56, %rsp
 
         movq 8(%rbp),%r10       // Grab the return pointer.
         incq %r10               // Skip past the `ret` in our parent frame