Give the GNU and AEABI functions the right ABIs

A buildbot reported a failure in a hardfp build, related to ABI: the
test was calling __mulsf3 and passing arguments in s0/s1, but the code
inside __mulsf3 was reading them out of r0/r1.

The ABI using GPRs is correct for __aeabi_fmul, but not for __mulsf3,
which takes float arguments in accordance with whatever the normal ABI
is. So in hardfp, the two functions behave differently.

The obvious question is why anyone is linking this function in to a
hardfp build in the first place - surely in a hardfp context clients
would just use a vmul instruction instead of calling either of these
entry points? But there seems to be no provision in
builtins/CMakeLists.txt for leaving things out of hardfp builds. The
generic __mulsf3.c is still included in a hardfp builtins library. So
I've stuck with those basic premises, and just corrected my
replacement functions to get the ABIs right.

Author: statham-arm

compiler-rt/lib/builtins/arm/divsf3.S

@@ -19,9 +19,19 @@
   .text
   .p2align 2
 
-DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_fdiv, __divsf3)
-
+#if __ARM_PCS_VFP
 DEFINE_COMPILERRT_FUNCTION(__divsf3)
+  push {r4, lr}
+  vmov r0, s0
+  vmov r1, s1
+  bl __aeabi_fdiv
+  vmov s0, r0
+  pop {r4, pc}
+#else
+DEFINE_COMPILERRT_FUNCTION_ALIAS(__divsf3, __aeabi_fdiv)
+#endif
+
+DEFINE_COMPILERRT_FUNCTION(__aeabi_fdiv)
   // Extract the exponents of the inputs into r2 and r3, occupying bits 16-23
   // of each register so that there will be space lower down to store extra
   // data without exponent arithmetic carrying into it. In the process, check
@@ -603,6 +613,6 @@ LOCAL_LABEL(tab):
   .byte 0x82  // input [0xfc0000,0xfdffff], candidate outputs [0x82,0x82]
   .byte 0x81  // input [0xfe0000,0xffffff], candidate outputs [0x80,0x81]
 
-END_COMPILERRT_FUNCTION(__divsf3)
+END_COMPILERRT_FUNCTION(__aeabi_fdiv)
 
 NO_EXEC_STACK_DIRECTIVE

compiler-rt/lib/builtins/arm/mulsf3.S

@@ -19,9 +19,19 @@
   .text
   .p2align 2
 
-DEFINE_AEABI_FUNCTION_ALIAS(__aeabi_fmul, __mulsf3)
-
+#if __ARM_PCS_VFP
 DEFINE_COMPILERRT_FUNCTION(__mulsf3)
+  push {r4, lr}
+  vmov r0, s0
+  vmov r1, s1
+  bl __aeabi_fmul
+  vmov s0, r0
+  pop {r4, pc}
+#else
+DEFINE_COMPILERRT_FUNCTION_ALIAS(__mulsf3, __aeabi_fmul)
+#endif
+
+DEFINE_COMPILERRT_FUNCTION(__aeabi_fmul)
 
   // Check if either input exponent is 00 or FF (i.e. not a normalized number),
   // and if so, branch out of line. If we don't branch out of line, then we've
@@ -304,6 +314,6 @@ LOCAL_LABEL(inf_NaN):
   orreq   r0, r0, #0x400000        // otherwise, set the 'quiet NaN' bit
   bx      lr                       // and return
 
-END_COMPILERRT_FUNCTION(__mulsf3)
+END_COMPILERRT_FUNCTION(__aeabi_fmul)
 
 NO_EXEC_STACK_DIRECTIVE