[libcpu][cortex-a] Improve FPU stack initialization implementation

Co-authored-by: BernardXiong <1241087+BernardXiong@users.noreply.github.com>
This commit is contained in:
copilot-swe-agent[bot]
2025-11-11 01:24:47 +00:00
parent 3241912ee8
commit 4dc072c04c
2 changed files with 17 additions and 1 deletions

View File

@@ -72,6 +72,11 @@ struct rt_hw_stack
#define E_Bit (1<<9)
#define J_Bit (1<<24)
/* VFP/NEON register count for FPU context */
#ifndef VFP_DATA_NR
#define VFP_DATA_NR 64 /* 32 double-precision registers = 64 words */
#endif
#ifdef RT_USING_SMP
typedef union {
unsigned long slock;

View File

@@ -31,6 +31,9 @@ rt_uint8_t *rt_hw_stack_init(void *tentry, void *parameter,
rt_uint8_t *stack_addr, void *texit)
{
rt_uint32_t *stk;
#ifdef RT_USING_FPU
rt_uint32_t i;
#endif
stack_addr += sizeof(rt_uint32_t);
stack_addr = (rt_uint8_t *)RT_ALIGN_DOWN((rt_uint32_t)stack_addr, 8);
@@ -61,7 +64,15 @@ rt_uint8_t *rt_hw_stack_init(void *tentry, void *parameter,
*(--stk) = 0; /* user sp*/
#endif
#ifdef RT_USING_FPU
*(--stk) = 0; /* not use fpu*/
/* FPU context initialization matches context_gcc.S restore order:
* Stack layout (high to low): FPEXC -> FPSCR -> D16-D31 -> D0-D15
*/
for (i = 0; i < VFP_DATA_NR; i++)
{
*(--stk) = 0; /* Initialize D0-D31 (64 words for 32 double regs) */
}
*(--stk) = 0; /* FPSCR: Floating-Point Status and Control Register */
*(--stk) = 0x40000000; /* FPEXC: Enable FPU (bit 30 = EN) */
#endif
/* return task's current stack address */