/* SPDX-License-Identifier: MIT */
/* Copyright (c) 2021-2022 The SRS Authors */

/* If user disable the ASM, such as avoiding bugs in ASM, donot compile it. */
#if !defined(MD_ST_NO_ASM)

#if defined(__aarch64__)

    /****************************************************************/
    /* See https://developer.arm.com/documentation/102374/0100/Function-calls */
    /* See https://developer.arm.com/documentation/102374/0100/Procedure-Call-Standard */
    /* See https://github.com/ARM-software/abi-aa/blob/main/aapcs64/aapcs64.rst#machine-registers */
    /* See https://wiki.cdot.senecacollege.ca/wiki/AArch64_Register_and_Instruction_Quick_Start */
    /* See https://chromium.googlesource.com/native_client/nacl-glibc/+/glibc-2.21/sysdeps/aarch64/__longjmp.S */
    /* See https://chromium.googlesource.com/native_client/nacl-glibc/+/glibc-2.21/sysdeps/aarch64/setjmp.S */

    /* The called routine is expected to preserve r19-r28 *** These registers are generally
        safe to use in your program. */
    #define JB_X19            0
    #define JB_X20            1
    #define JB_X21            2
    #define JB_X22            3
    #define JB_X23            4
    #define JB_X24            5
    #define JB_X25            6
    #define JB_X26            7
    #define JB_X27            8
    #define JB_X28            9
    /* r29 and r30 are used as the frame register and link register (avoid) */
    #define JB_X29           10
    #define JB_LR            11
    /* Register '31' is one of two registers depending on the instruction context:
        For instructions dealing with the stack, it is the stack pointer, named rsp */
    #define JB_SP		 13

    /* FP registers */
    #define JB_D8		 14
    #define JB_D9		 15
    #define JB_D10		 16
    #define JB_D11		 17
    #define JB_D12		 18
    #define JB_D13		 19
    #define JB_D14		 20
    #define JB_D15		 21

    	.file "md.S"
    	.text

    	/* _st_md_cxt_save(__jmp_buf env) */
    	.globl _st_md_cxt_save
    	.type _st_md_cxt_save, %function
    	.align 4
    _st_md_cxt_save:
    	stp	x19, x20, [x0, #JB_X19<<3]
    	stp	x21, x22, [x0, #JB_X21<<3]
    	stp	x23, x24, [x0, #JB_X23<<3]
    	stp	x25, x26, [x0, #JB_X25<<3]
    	stp	x27, x28, [x0, #JB_X27<<3]
    	stp	x29, x30, [x0, #JB_X29<<3]

    	stp	 d8,  d9, [x0, #JB_D8<<3]
    	stp	d10, d11, [x0, #JB_D10<<3]
    	stp	d12, d13, [x0, #JB_D12<<3]
    	stp	d14, d15, [x0, #JB_D14<<3]
    	mov	x2,  sp
    	str	x2,  [x0, #JB_SP<<3]

    	mov	x0, #0
    	ret
    	.size _st_md_cxt_save, .-_st_md_cxt_save

    /****************************************************************/

    	/* _st_md_cxt_restore(__jmp_buf env, int val) */
    	.globl _st_md_cxt_restore
    	.type _st_md_cxt_restore, %function
    	.align 4
    _st_md_cxt_restore:
    	ldp	x19, x20, [x0, #JB_X19<<3]
    	ldp	x21, x22, [x0, #JB_X21<<3]
    	ldp	x23, x24, [x0, #JB_X23<<3]
    	ldp	x25, x26, [x0, #JB_X25<<3]
    	ldp	x27, x28, [x0, #JB_X27<<3]

    	ldp	x29, x30, [x0, #JB_X29<<3]

    	ldp	 d8,  d9, [x0, #JB_D8<<3]
    	ldp	d10, d11, [x0, #JB_D10<<3]
    	ldp	d12, d13, [x0, #JB_D12<<3]
    	ldp	d14, d15, [x0, #JB_D14<<3]

    	ldr	x5, [x0, #JB_SP<<3]
    	mov	sp, x5

    	cmp	x1, #0
    	mov	x0, #1
    	csel	x0, x1, x0, ne
    	/* Use br instead of ret because ret is guaranteed to mispredict */
    	br	x30
    	.size _st_md_cxt_restore, .-_st_md_cxt_restore

    /****************************************************************/










#elif defined(__arm__)

    /****************************************************************/
    /* https://github.com/ossrs/srs/issues/1282#issuecomment-445539513 */

    /* Register list for a ldm/stm instruction to load/store
       the general registers from a __jmp_buf.  */
    # define JMP_BUF_REGLIST	{v1-v6, sl, fp, sp, lr}

    	.file "md.S"
    	.text

    	/* _st_md_cxt_save(__jmp_buf env) */
    	.globl _st_md_cxt_save
    	.type _st_md_cxt_save, %function
    	.align 2
    _st_md_cxt_save:
    	mov	ip, r0

    	/* Save registers */
    	stmia	ip!, JMP_BUF_REGLIST

    #ifdef __VFP_FP__
    	/* Store the VFP registers.  */
    	/* Following instruction is vstmia ip!, {d8-d15}.  */
    	stc	p11, cr8, [ip], #64
    #endif

    #ifdef __IWMMXT__
    	/* Save the call-preserved iWMMXt registers.  */
    	/* Following instructions are wstrd wr10, [ip], #8 (etc.)  */
    	stcl	p1, cr10, [r12], #8
    	stcl	p1, cr11, [r12], #8
    	stcl	p1, cr12, [r12], #8
    	stcl	p1, cr13, [r12], #8
    	stcl	p1, cr14, [r12], #8
    	stcl	p1, cr15, [r12], #8
    #endif

    	mov	r0, #0
    	bx	lr

    	.size _st_md_cxt_save, .-_st_md_cxt_save

    /****************************************************************/

    	/* _st_md_cxt_restore(__jmp_buf env, int val) */
    	.globl _st_md_cxt_restore
    	.type _st_md_cxt_restore, %function
    	.align 2
    _st_md_cxt_restore:
    	mov	ip, r0

    	/* Restore registers */
    	ldmia	ip!, JMP_BUF_REGLIST

    #ifdef __VFP_FP__
    	/* Restore the VFP registers.  */
    	/* Following instruction is vldmia ip!, {d8-d15}.  */
    	ldc	p11, cr8, [r12], #64
    #endif

    #ifdef __IWMMXT__
    	/* Restore the call-preserved iWMMXt registers.  */
    	/* Following instructions are wldrd wr10, [ip], #8 (etc.)  */
    	ldcl	p1, cr10, [r12], #8
    	ldcl	p1, cr11, [r12], #8
    	ldcl	p1, cr12, [r12], #8
    	ldcl	p1, cr13, [r12], #8
    	ldcl	p1, cr14, [r12], #8
    	ldcl	p1, cr15, [r12], #8
    #endif

    	movs	r0, r1		/* get the return value in place */
    	moveq	r0, #1		/* can't let setjmp() return zero! */
    	bx	lr

    	.size _st_md_cxt_restore, .-_st_md_cxt_restore

    /****************************************************************/





#elif defined(__riscv)

    /****************************************************************/
    /*
     * Internal __jmp_buf layout
     * riscv-asm: https://github.com/riscv/riscv-asm-manual/blob/master/riscv-asm.md
     */
    #define JB_SP  0    /* A0, SP, Stack pointer */
    #define JB_RA  1    /* RA, Return address */
    #define JB_FP  2    /* FP/S0 Frame pointer */
    #define JB_S1  3    /* S1 Saved register*/
    #define JB_S2  4    /* S2-S11,  Saved register */
    #define JB_S3  5    /* S2-S11,  Saved register */
    #define JB_S4  6    /* S2-S11,  Saved register */
    #define JB_S5  7    /* S2-S11,  Saved register */
    #define JB_S6  8    /* S2-S11,  Saved register */
    #define JB_S7  9    /* S2-S11,  Saved register */
    #define JB_S8  10    /* S2-S11,  Saved register */
    #define JB_S9  11   /* S2-S11,  Saved register */
    #define JB_S10 12   /* S2-S11,  Saved register */
    #define JB_S11 13   /* S2-S11,  Saved register */


    	.file "md_linux.S"
    	.text

    	/* _st_md_cxt_save(__jmp_buf env) */ /* The env is $a0, https://en.wikipedia.org/wiki/RISC-V#Register_sets */
    	.globl _st_md_cxt_save
    	.type _st_md_cxt_save, %function
    	.align 2
    _st_md_cxt_save:
	sd    sp,  JB_SP  * 8(a0)
	sd    ra,  JB_RA  * 8(a0)
	sd    s0,  JB_FP  * 8(a0)
	sd    s1,  JB_S1  * 8(a0)
	sd    s2,  JB_S2  * 8(a0)
	sd    s3,  JB_S3  * 8(a0)
	sd    s4,  JB_S4  * 8(a0)
	sd    s5,  JB_S5  * 8(a0)
	sd    s6,  JB_S6  * 8(a0)
	sd    s7,  JB_S7  * 8(a0)
	sd    s8,  JB_S8  * 8(a0)
	sd    s9,  JB_S9  * 8(a0)
	sd    s10, JB_S10 * 8(a0)
	sd    s11, JB_S11 * 8(a0)
	li    a0,  0
	jr    ra
    	.size _st_md_cxt_save, .-_st_md_cxt_save

    /****************************************************************/

    	/* _st_md_cxt_restore(__jmp_buf env, int val) */
    	.globl _st_md_cxt_restore
    	.type _st_md_cxt_restore, %function
    	.align 2
    _st_md_cxt_restore:
	ld        sp,  JB_SP  * 8(a0)
	ld        ra,  JB_RA  * 8(a0)
	ld        s0,  JB_FP  * 8(a0)
	ld        s1,  JB_S1  * 8(a0)
	ld        s2,  JB_S2  * 8(a0)
	ld        s3,  JB_S3  * 8(a0)
	ld        s4,  JB_S4  * 8(a0)
	ld        s5,  JB_S5  * 8(a0)
	ld        s6,  JB_S6  * 8(a0)
	ld        s7,  JB_S7  * 8(a0)
	ld        s8,  JB_S8  * 8(a0)
	ld        s9,  JB_S9  * 8(a0)
	ld        s10, JB_S10 * 8(a0)
	ld        s11, JB_S11 * 8(a0)
	li        a0,  1
	jr        ra
    	.size _st_md_cxt_restore, .-_st_md_cxt_restore

    /****************************************************************/






#elif defined(__mips64)

    /****************************************************************/
    /* For MIPS64, see https://s3-eu-west-1.amazonaws.com/downloads-mips/documents/MIPS_Architecture_MIPS64_InstructionSet_%20AFP_P_MD00087_06.05.pdf */

    /*
     * Internal __jmp_buf layout
     */
    #define JB_SP  0    /* Stack pointer */
    #define JB_RA  11   /* Return address */
    #define JB_GP  1    /* Global pointer */
    #define JB_S0  3    /* S0-S7, Saved temporaries */
    #define JB_S1  4    /* S0-S7, Saved temporaries */
    #define JB_S2  5    /* S0-S7, Saved temporaries */
    #define JB_S3  6    /* S0-S7, Saved temporaries */
    #define JB_S4  7    /* S0-S7, Saved temporaries */
    #define JB_S5  8    /* S0-S7, Saved temporaries */
    #define JB_S6  9    /* S0-S7, Saved temporaries */
    #define JB_S7  10   /* S0-S7, Saved temporaries */
    #define JB_FP  2    /* FP/S8 Frame pointer */

    	.file "md_linux.S"
    	.text

    	/* _st_md_cxt_save(__jmp_buf env) */ /* The env is $a0, https://en.wikipedia.org/wiki/MIPS_architecture#Calling_conventions */
    	.globl _st_md_cxt_save
    	.type _st_md_cxt_save, %function
    	.align 2
    _st_md_cxt_save:
        sd $sp,  0($a0) /* Save sp to env[0],  *(long*)($a0+ 0) =sp */
        sd $ra,  8($a0) /* Save ra to env[1],  *(long*)($a0+ 8)=ra, the return address, https://chortle.ccsu.edu/AssemblyTutorial/Chapter-26/ass26_4.html */
        sd $gp, 16($a0) /* Save gp to env[2],  *(long*)($a0+16) =gp */
        sd $s0, 24($a0) /* Save s0 to env[3],  *(long*)($a0+24)=s0 */
        sd $s1, 32($a0) /* Save s1 to env[4],  *(long*)($a0+32)=s1 */
        sd $s2, 40($a0) /* Save s2 to env[5],  *(long*)($a0+40)=s2 */
        sd $s3, 48($a0) /* Save s3 to env[6],  *(long*)($a0+48)=s3 */
        sd $s4, 56($a0) /* Save s4 to env[7],  *(long*)($a0+56)=s4 */
        sd $s5, 64($a0) /* Save s5 to env[8],  *(long*)($a0+64)=s5 */
        sd $s6, 72($a0) /* Save s6 to env[9],  *(long*)($a0+72)=s6 */
        sd $s7, 80($a0) /* Save s7 to env[10], *(long*)($a0+80)=s7 */
        sd $fp, 88($a0) /* Save fp to env[11], *(long*)($a0+88) =fp */
        li $v0, 0       /* Set return value to 0 */
        jr $ra          /* Return */

    	.size _st_md_cxt_save, .-_st_md_cxt_save

    /****************************************************************/

    	/* _st_md_cxt_restore(__jmp_buf env, int val) */
    	.globl _st_md_cxt_restore
    	.type _st_md_cxt_restore, %function
    	.align 2
    _st_md_cxt_restore:
        ld $sp,  0($a0) /* Load sp from env[0],  sp=*(long*)($a0+ 0)  */
        ld $ra,  8($a0) /* Load sp from env[1],  ra=*(long*)($a0+ 8), the saved return address */
        ld $gp, 16($a0) /* Load sp from env[2],  gp=*(long*)($a0+16)  */
        ld $s0, 24($a0) /* Load sp from env[3],  s0=*(long*)($a0+24) */
        ld $s1, 32($a0) /* Load sp from env[4],  s1=*(long*)($a0+32) */
        ld $s2, 40($a0) /* Load sp from env[5],  s2=*(long*)($a0+40) */
        ld $s3, 48($a0) /* Load sp from env[6],  s3=*(long*)($a0+48) */
        ld $s4, 56($a0) /* Load sp from env[7],  s4=*(long*)($a0+56) */
        ld $s5, 64($a0) /* Load sp from env[8],  s5=*(long*)($a0+64) */
        ld $s6, 72($a0) /* Load sp from env[9],  s6=*(long*)($a0+72) */
        ld $s7, 80($a0) /* Load sp from env[10], s7=*(long*)($a0+80) */
        ld $fp, 88($a0) /* Load sp from env[2],  fp=*(long*)($a0+88)  */
        li $v0, 1       /* Set return value to 1 */
        jr $ra          /* Return to the saved return address */

    	.size _st_md_cxt_restore, .-_st_md_cxt_restore

    /****************************************************************/






#elif defined(__mips__)

    /****************************************************************/
    /* For MIPS32, see https://s3-eu-west-1.amazonaws.com/downloads-mips/documents/MD00565-2B-MIPS32-QRC-01.01.pdf */

    /*
     * Internal __jmp_buf layout
     */
    #define JB_SP  0    /* Stack pointer */
    #define JB_RA  11   /* Return address */
    #define JB_GP  1    /* Global pointer */
    #define JB_S0  3    /* S0-S7, Saved temporaries */
    #define JB_S1  4    /* S0-S7, Saved temporaries */
    #define JB_S2  5    /* S0-S7, Saved temporaries */
    #define JB_S3  6    /* S0-S7, Saved temporaries */
    #define JB_S4  7    /* S0-S7, Saved temporaries */
    #define JB_S5  8    /* S0-S7, Saved temporaries */
    #define JB_S6  9    /* S0-S7, Saved temporaries */
    #define JB_S7  10   /* S0-S7, Saved temporaries */
    #define JB_FP  2    /* FP/S8 Frame pointer */

    	.file "md_linux.S"
    	.text

    	/* _st_md_cxt_save(__jmp_buf env) */ /* The env is $a0, https://en.wikipedia.org/wiki/MIPS_architecture#Calling_conventions */
    	.globl _st_md_cxt_save
    	.type _st_md_cxt_save, %function
    	.align 2
    _st_md_cxt_save:
        sw $sp, 0($a0)  /* Save sp to env[0],  *(long*)($a0+0) =sp */
        sw $ra, 4($a0)  /* Save ra to env[1],  *(long*)($a0+4)=ra, the return address, https://chortle.ccsu.edu/AssemblyTutorial/Chapter-26/ass26_4.html */
        sw $gp, 8($a0)  /* Save gp to env[2],  *(long*)($a0+8) =gp */
        sw $s0, 12($a0) /* Save s0 to env[3],  *(long*)($a0+12)=s0 */
        sw $s1, 16($a0) /* Save s1 to env[4],  *(long*)($a0+16)=s1 */
        sw $s2, 20($a0) /* Save s2 to env[5],  *(long*)($a0+20)=s2 */
        sw $s3, 24($a0) /* Save s3 to env[6],  *(long*)($a0+24)=s3 */
        sw $s4, 28($a0) /* Save s4 to env[7],  *(long*)($a0+28)=s4 */
        sw $s5, 32($a0) /* Save s5 to env[8],  *(long*)($a0+32)=s5 */
        sw $s6, 36($a0) /* Save s6 to env[9],  *(long*)($a0+36)=s6 */
        sw $s7, 40($a0) /* Save s7 to env[10], *(long*)($a0+40)=s7 */
        sw $fp, 44($a0) /* Save fp to env[11], *(long*)($a0+44) =fp */
        li $v0, 0       /* Set return value to 0 */
        jr $ra          /* Return */

    	.size _st_md_cxt_save, .-_st_md_cxt_save

    /****************************************************************/

    	/* _st_md_cxt_restore(__jmp_buf env, int val) */
    	.globl _st_md_cxt_restore
    	.type _st_md_cxt_restore, %function
    	.align 2
    _st_md_cxt_restore:
        lw $sp, 0($a0)  /* Load sp from env[0],  sp=*(long*)($a0+0)  */
        lw $ra, 4($a0)  /* Load sp from env[1],  ra=*(long*)($a0+4), the saved return address */
        lw $gp, 8($a0)  /* Load sp from env[2],  gp=*(long*)($a0+8)  */
        lw $s0, 12($a0) /* Load sp from env[3],  s0=*(long*)($a0+12) */
        lw $s1, 16($a0) /* Load sp from env[4],  s1=*(long*)($a0+16) */
        lw $s2, 20($a0) /* Load sp from env[5],  s2=*(long*)($a0+20) */
        lw $s3, 24($a0) /* Load sp from env[6],  s3=*(long*)($a0+24) */
        lw $s4, 28($a0) /* Load sp from env[7],  s4=*(long*)($a0+28) */
        lw $s5, 32($a0) /* Load sp from env[8],  s5=*(long*)($a0+32) */
        lw $s6, 36($a0) /* Load sp from env[9],  s6=*(long*)($a0+36) */
        lw $s7, 40($a0) /* Load sp from env[10], s7=*(long*)($a0+40) */
        lw $fp, 44($a0) /* Load sp from env[2],  fp=*(long*)($a0+44)  */
        li $v0, 1       /* Set return value to 1 */
        jr $ra          /* Return to the saved return address */

    	.size _st_md_cxt_restore, .-_st_md_cxt_restore

    /****************************************************************/










#elif defined(__loongarch64)

    /****************************************************************/

    /*
     * Internal __jmp_buf layout
     */
    #define JB_SP  0    /* R3, SP, Stack pointer */
    #define JB_RA  1    /* R1, RA, Return address */
    #define JB_FP  2    /* FP/R22 Frame pointer */
    #define JB_S0  3    /* R23-R31, S0-S8, Subroutine register variable */
    #define JB_S1  4    /* R23-R31, S0-S8, Subroutine register variable */
    #define JB_S2  5    /* R23-R31, S0-S8, Subroutine register variable */
    #define JB_S3  6    /* R23-R31, S0-S8, Subroutine register variable */
    #define JB_S4  7    /* R23-R31, S0-S8, Subroutine register variable */
    #define JB_S5  8    /* R23-R31, S0-S8, Subroutine register variable */
    #define JB_S6  9    /* R23-R31, S0-S8, Subroutine register variable */
    #define JB_S7  10   /* R23-R31, S0-S8, Subroutine register variable */
    #define JB_S8  11   /* R23-R31, S0-S8, Subroutine register variable */

    	.file "md_linux.S"
    	.text

    	/* _st_md_cxt_save(__jmp_buf env) */ /* The env is $r4, https://github.com/ossrs/state-threads/issues/24#porting */
    	.globl _st_md_cxt_save
    	.type _st_md_cxt_save, %function
    	.align 2
    _st_md_cxt_save:
        st.d	$r3, $r4, 0           /* Save sp to env[0],  *(long*)($r4+0) = sp */
        st.d	$r1, $r4, 8           /* Save ra to env[1],  *(long*)($r4+8) = r1 */
        st.d	$r22, $r4, 16         /* Save fp to env[2],  *(long*)($r4+16) = r22 */
        st.d	$r23, $r4, 24         /* Save r23 to env[3],  *(long*)($r4+24) = r23 */
        st.d	$r24, $r4, 32         /* Save r24 to env[4],  *(long*)($r4+32) = r24 */
        st.d	$r25, $r4, 40         /* Save r25 to env[5],  *(long*)($r4+40) = r25 */
        st.d	$r26, $r4, 48         /* Save r26 to env[6],  *(long*)($r4+48) = r26 */
        st.d	$r27, $r4, 56         /* Save r27 to env[7],  *(long*)($r4+56) = r27 */
        st.d	$r28, $r4, 64         /* Save r28 to env[8],  *(long*)($r4+64) = r28 */
        st.d	$r29, $r4, 72         /* Save r29 to env[9],  *(long*)($r4+72) = r29 */
        st.d	$r30, $r4, 80         /* Save r30 to env[10],  *(long*)($r4+80) = r30 */
        st.d	$r31, $r4, 88         /* Save r31 to env[11],  *(long*)($r4+88) = r31 */
        addi.w	$r12, $r0, 0          /* Set return value to 0 */
        move    $r4, $r12             /* Set return value to 0 */
        jirl	$r0, $r1, 0           /* Return */

    	.size _st_md_cxt_save, .-_st_md_cxt_save

    /****************************************************************/

    	/* _st_md_cxt_restore(__jmp_buf env, int val) */
    	.globl _st_md_cxt_restore
    	.type _st_md_cxt_restore, %function
    	.align 2
    _st_md_cxt_restore:
        ld.d	$r3, $r4, 0          /* Load sp from env[0],  sp=*(long*)($r4+0) */
        ld.d	$r1, $r4, 8          /* Load ra from env[1],  r1=*(long*)($r4+8) */
        ld.d	$r22, $r4, 16        /* Load fp from env[2],  r22=*(long*)($r4+16) */
        ld.d	$r23, $r4, 24        /* Load r23 from env[3],  r23=*(long*)($r4+24) */
        ld.d	$r24, $r4, 32        /* Load r24 from env[4],  r24=*(long*)($r4+32) */
        ld.d	$r25, $r4, 40        /* Load r25 from env[5],  r25=*(long*)($r4+40) */
        ld.d	$r26, $r4, 48        /* Load r26 from env[6],  r26=*(long*)($r4+48) */
        ld.d	$r27, $r4, 56        /* Load r27 from env[7],  r27=*(long*)($r4+56) */
        ld.d	$r28, $r4, 64        /* Load r28 from env[8],  r28=*(long*)($r4+64) */
        ld.d	$r29, $r4, 72        /* Load r29 from env[9],  r29=*(long*)($r4+72) */
        ld.d	$r30, $r4, 80        /* Load r30 from env[10],  r30=*(long*)($r4+80) */
        ld.d	$r31, $r4, 88        /* Load r31 from env[11],  r31=*(long*)($r4+88) */
        addi.w	$r12, $r0, 1         /* Set return value to 1 */
        move    $r4, $r12            /* Set return value to 1 */
        jirl	$r0, $r1, 0          /* Return to the saved return address */

    	.size _st_md_cxt_restore, .-_st_md_cxt_restore

    /****************************************************************/

#endif

#endif