Fixup divider save_restore for floating point too; improve tests (#405)

- The divider state needs to be saved for __aeabi_ddiv, __aeabi_fdiv, __aeabi_dtan and __aeabi_ftan or they won't work in interrupts *(probably not used much youd hope), or on an RTOS context switch
 - Refactored code out for the integer and floating point cases
 - Improved the floating point 'tests' in passing to check more return values against GCC implementations
 - Added floating point usage to the IRQ nesting test case

src/rp2_common/pico_float/float_aeabi.S

@@ -6,6 +6,7 @@
 
 #include "pico/asm_helper.S"
 #include "pico/bootrom/sf_table.h"
+#include "hardware/divider_helper.S"
 
 __pre_init __aeabi_float_init, 00020
 
@@ -104,16 +105,16 @@ __check_nan_f2:
 .endm
 
 
-# note generally each function is in a separate section unless there is fall thru or branching between them
-# note fadd, fsub, fmul, fdiv are so tiny and just defer to rom so are lumped together so they can share constant pool
+// note generally each function is in a separate section unless there is fall thru or branching between them
+// note fadd, fsub, fmul, fdiv are so tiny and just defer to rom so are lumped together so they can share constant pool
 
-# note functions are word aligned except where they are an odd number of linear instructions
+// note functions are word aligned except where they are an odd number of linear instructions
 
 // float FUNC_NAME(__aeabi_fadd)(float, float)         single-precision addition
 float_wrapper_section __aeabi_farithmetic
 // float FUNC_NAME(__aeabi_frsub)(float x, float y)    single-precision reverse subtraction, y - x
 
-# frsub first because it is the only one that needs alignment
+// frsub first because it is the only one that needs alignment
 .align 2
 wrapper_func __aeabi_frsub
     eors r0, r1
@@ -146,7 +147,30 @@ wrapper_func_f2 __aeabi_fdiv
     b fdiv_fsub_nan_helper
 1:
 #endif
+#if !PICO_DIVIDER_DISABLE_INTERRUPTS
+    // to support IRQ usage (or context switch) we must save/restore divider state around call if state is dirty
+    ldr r2, =(SIO_BASE)
+    ldr r3, [r2, #SIO_DIV_CSR_OFFSET]
+    lsrs r3, #SIO_DIV_CSR_DIRTY_SHIFT_FOR_CARRY
+    bcs fdiv_save_state
+#else
+    // to avoid worrying about IRQs (or context switches), simply disable interrupts around call
+    push {r4, lr}
+    mrs r4, PRIMASK
+    cpsid i
+    bl fdiv_shim_call
+    msr PRIMASK, r4
+    pop {r4, pc}
+#endif
+fdiv_shim_call:
     table_tail_call SF_TABLE_FDIV
+#if !PICO_DIVIDER_DISABLE_INTERRUPTS
+fdiv_save_state:
+    save_div_state_and_lr
+    bl fdiv_shim_call
+    ldr r2, =(SIO_BASE)
+    restore_div_state_and_return
+#endif
 
 fdiv_fsub_nan_helper:
 #if PICO_FLOAT_PROPAGATE_NANS
@@ -689,10 +713,33 @@ wrapper_func tanf
     lsls r1, r0, #1
     lsrs r1, #24
     cmp r1, #127 + 7
-    bge 1f
-2:
+    bge ftan_out_of_range
+ftan_in_range:
+#if !PICO_DIVIDER_DISABLE_INTERRUPTS
+    // to support IRQ usage (or context switch) we must save/restore divider state around call if state is dirty
+    ldr r2, =(SIO_BASE)
+    ldr r3, [r2, #SIO_DIV_CSR_OFFSET]
+    lsrs r3, #SIO_DIV_CSR_DIRTY_SHIFT_FOR_CARRY
+    bcs ftan_save_state
+#else
+    // to avoid worrying about IRQs (or context switches), simply disable interrupts around call
+    push {r4, lr}
+    mrs r4, PRIMASK
+    cpsid i
+    bl ftan_shim_call
+    msr PRIMASK, r4
+    pop {r4, pc}
+#endif
+ftan_shim_call:
     table_tail_call SF_TABLE_FTAN
-1:
+#if !PICO_DIVIDER_DISABLE_INTERRUPTS
+ftan_save_state:
+    save_div_state_and_lr
+    bl ftan_shim_call
+    ldr r2, =(SIO_BASE)
+    restore_div_state_and_return
+#endif
+ftan_out_of_range:
 #if PICO_FLOAT_PROPAGATE_NANS
     // also check for infinites
     cmp r1, #255
@@ -709,7 +756,7 @@ wrapper_func tanf
     bl remainderf
     pop {r1}
     mov lr, r1
-    b 2b
+    b ftan_in_range
 
 float_wrapper_section atan2f
 wrapper_func_f2 atan2f