1
;Check 5.5 Parameter Passing --> Stage C --> C.4 statement, when NSAA is not
4
; Our purpose: make NSAA != SP, and only after start to use GPRs.
6
;Co-Processor register candidates may be either in VFP or in stack, so after
7
;all VFP are allocated, stack is used. We can use stack without GPR allocation
8
;in that case, passing 9 f64 params, for example.
9
;First eight params goes to d0-d7, ninth one goes to the stack.
10
;Now, as 10th parameter, we pass i32, and it must go to R0.
12
;5.5 Parameter Passing, Stage C:
14
;C.2.cp If the argument is a CPRC then any co-processor registers in that class
15
;that are unallocated are marked as unavailable. The NSAA is adjusted upwards
16
;until it is correctly aligned for the argument and the argument is copied to
17
;the memory at the adjusted NSAA. The NSAA is further incremented by the size
18
;of the argument. The argument has now been allocated.
20
;C.4 If the size in words of the argument is not more than r4 minus NCRN, the
21
;argument is copied into core registers, starting at the NCRN. The NCRN is
22
;incremented by the number of registers used. Successive registers hold the
23
;parts of the argument they would hold if its value were loaded into those
24
;registers from memory using an LDM instruction. The argument has now been
27
;What is actually checked here:
28
;Here we check that i32 param goes to r0.
30
;Current test-case was produced with command:
31
;arm-linux-gnueabihf-clang -mcpu=cortex-a9 params-to-GPR.c -S -O1 -emit-llvm
35
;void fooUseI32(unsigned);
37
;void foo(long double p0,
51
; foo( 1,2,3,4,5,6,7,8,9, 43 );
54
;RUN: llc -mtriple=thumbv7-linux-gnueabihf -float-abi=hard < %s | FileCheck %s
64
define void @foo(double %p0, ; --> D0
72
double %p8, ; --> Stack
73
i32 %p9) #0 { ; --> R0, not Stack+8
75
tail call void @fooUseI32(i32 %p9)
79
declare void @fooUseI32(i32)
81
define void @doFoo() {
83
tail call void @foo(double 23.0, ; --> D0
91
double 23.8, ; --> Stack
92
i32 43) ; --> R0, not Stack+8