OVMS3-idf/components/esp32/include/xtensa/coreasm.h
2016-08-17 23:08:22 +08:00

939 lines
27 KiB
C
Executable file

/*
* xtensa/coreasm.h -- assembler-specific definitions that depend on CORE configuration
*
* Source for configuration-independent binaries (which link in a
* configuration-specific HAL library) must NEVER include this file.
* It is perfectly normal, however, for the HAL itself to include this file.
*
* This file must NOT include xtensa/config/system.h. Any assembler
* header file that depends on system information should likely go
* in a new systemasm.h (or sysasm.h) header file.
*
* NOTE: macro beqi32 is NOT configuration-dependent, and is placed
* here until we have a proper configuration-independent header file.
*/
/* $Id: //depot/rel/Eaglenest/Xtensa/OS/include/xtensa/coreasm.h#3 $ */
/*
* Copyright (c) 2000-2014 Tensilica Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#ifndef XTENSA_COREASM_H
#define XTENSA_COREASM_H
/*
* Tell header files this is assembly source, so they can avoid non-assembler
* definitions (eg. C types etc):
*/
#ifndef _ASMLANGUAGE /* conditionalize to avoid cpp warnings (3rd parties might use same macro) */
#define _ASMLANGUAGE
#endif
#include <xtensa/config/core.h>
#include <xtensa/config/specreg.h>
#include <xtensa/config/system.h>
/*
* Assembly-language specific definitions (assembly macros, etc.).
*/
/*----------------------------------------------------------------------
* find_ms_setbit
*
* This macro finds the most significant bit that is set in <as>
* and return its index + <base> in <ad>, or <base> - 1 if <as> is zero.
* The index counts starting at zero for the lsbit, so the return
* value ranges from <base>-1 (no bit set) to <base>+31 (msbit set).
*
* Parameters:
* <ad> destination address register (any register)
* <as> source address register
* <at> temporary address register (must be different than <as>)
* <base> constant value added to result (usually 0 or 1)
* On entry:
* <ad> = undefined if different than <as>
* <as> = value whose most significant set bit is to be found
* <at> = undefined
* no other registers are used by this macro.
* On exit:
* <ad> = <base> + index of msbit set in original <as>,
* = <base> - 1 if original <as> was zero.
* <as> clobbered (if not <ad>)
* <at> clobbered (if not <ad>)
* Example:
* find_ms_setbit a0, a4, a0, 0 -- return in a0 index of msbit set in a4
*/
.macro find_ms_setbit ad, as, at, base
#if XCHAL_HAVE_NSA
movi \at, 31+\base
nsau \as, \as // get index of \as, numbered from msbit (32 if absent)
sub \ad, \at, \as // get numbering from lsbit (0..31, -1 if absent)
#else /* XCHAL_HAVE_NSA */
movi \at, \base // start with result of 0 (point to lsbit of 32)
beqz \as, 2f // special case for zero argument: return -1
bltui \as, 0x10000, 1f // is it one of the 16 lsbits? (if so, check lower 16 bits)
addi \at, \at, 16 // no, increment result to upper 16 bits (of 32)
//srli \as, \as, 16 // check upper half (shift right 16 bits)
extui \as, \as, 16, 16 // check upper half (shift right 16 bits)
1: bltui \as, 0x100, 1f // is it one of the 8 lsbits? (if so, check lower 8 bits)
addi \at, \at, 8 // no, increment result to upper 8 bits (of 16)
srli \as, \as, 8 // shift right to check upper 8 bits
1: bltui \as, 0x10, 1f // is it one of the 4 lsbits? (if so, check lower 4 bits)
addi \at, \at, 4 // no, increment result to upper 4 bits (of 8)
srli \as, \as, 4 // shift right 4 bits to check upper half
1: bltui \as, 0x4, 1f // is it one of the 2 lsbits? (if so, check lower 2 bits)
addi \at, \at, 2 // no, increment result to upper 2 bits (of 4)
srli \as, \as, 2 // shift right 2 bits to check upper half
1: bltui \as, 0x2, 1f // is it the lsbit?
addi \at, \at, 2 // no, increment result to upper bit (of 2)
2: addi \at, \at, -1 // (from just above: add 1; from beqz: return -1)
//srli \as, \as, 1
1: // done! \at contains index of msbit set (or -1 if none set)
.if 0x\ad - 0x\at // destination different than \at ? (works because regs are a0-a15)
mov \ad, \at // then move result to \ad
.endif
#endif /* XCHAL_HAVE_NSA */
.endm // find_ms_setbit
/*----------------------------------------------------------------------
* find_ls_setbit
*
* This macro finds the least significant bit that is set in <as>,
* and return its index in <ad>.
* Usage is the same as for the find_ms_setbit macro.
* Example:
* find_ls_setbit a0, a4, a0, 0 -- return in a0 index of lsbit set in a4
*/
.macro find_ls_setbit ad, as, at, base
neg \at, \as // keep only the least-significant bit that is set...
and \as, \at, \as // ... in \as
find_ms_setbit \ad, \as, \at, \base
.endm // find_ls_setbit
/*----------------------------------------------------------------------
* find_ls_one
*
* Same as find_ls_setbit with base zero.
* Source (as) and destination (ad) registers must be different.
* Provided for backward compatibility.
*/
.macro find_ls_one ad, as
find_ls_setbit \ad, \as, \ad, 0
.endm // find_ls_one
/*----------------------------------------------------------------------
* floop, floopnez, floopgtz, floopend
*
* These macros are used for fast inner loops that
* work whether or not the Loops options is configured.
* If the Loops option is configured, they simply use
* the zero-overhead LOOP instructions; otherwise
* they use explicit decrement and branch instructions.
*
* They are used in pairs, with floop, floopnez or floopgtz
* at the beginning of the loop, and floopend at the end.
*
* Each pair of loop macro calls must be given the loop count
* address register and a unique label for that loop.
*
* Example:
*
* movi a3, 16 // loop 16 times
* floop a3, myloop1
* :
* bnez a7, end1 // exit loop if a7 != 0
* :
* floopend a3, myloop1
* end1:
*
* Like the LOOP instructions, these macros cannot be
* nested, must include at least one instruction,
* cannot call functions inside the loop, etc.
* The loop can be exited by jumping to the instruction
* following floopend (or elsewhere outside the loop),
* or continued by jumping to a NOP instruction placed
* immediately before floopend.
*
* Unlike LOOP instructions, the register passed to floop*
* cannot be used inside the loop, because it is used as
* the loop counter if the Loops option is not configured.
* And its value is undefined after exiting the loop.
* And because the loop counter register is active inside
* the loop, you can't easily use this construct to loop
* across a register file using ROTW as you might with LOOP
* instructions, unless you copy the loop register along.
*/
/* Named label version of the macros: */
.macro floop ar, endlabel
floop_ \ar, .Lfloopstart_\endlabel, .Lfloopend_\endlabel
.endm
.macro floopnez ar, endlabel
floopnez_ \ar, .Lfloopstart_\endlabel, .Lfloopend_\endlabel
.endm
.macro floopgtz ar, endlabel
floopgtz_ \ar, .Lfloopstart_\endlabel, .Lfloopend_\endlabel
.endm
.macro floopend ar, endlabel
floopend_ \ar, .Lfloopstart_\endlabel, .Lfloopend_\endlabel
.endm
/* Numbered local label version of the macros: */
#if 0 /*UNTESTED*/
.macro floop89 ar
floop_ \ar, 8, 9f
.endm
.macro floopnez89 ar
floopnez_ \ar, 8, 9f
.endm
.macro floopgtz89 ar
floopgtz_ \ar, 8, 9f
.endm
.macro floopend89 ar
floopend_ \ar, 8b, 9
.endm
#endif /*0*/
/* Underlying version of the macros: */
.macro floop_ ar, startlabel, endlabelref
.ifdef _infloop_
.if _infloop_
.err // Error: floop cannot be nested
.endif
.endif
.set _infloop_, 1
#if XCHAL_HAVE_LOOPS
loop \ar, \endlabelref
#else /* XCHAL_HAVE_LOOPS */
\startlabel:
addi \ar, \ar, -1
#endif /* XCHAL_HAVE_LOOPS */
.endm // floop_
.macro floopnez_ ar, startlabel, endlabelref
.ifdef _infloop_
.if _infloop_
.err // Error: floopnez cannot be nested
.endif
.endif
.set _infloop_, 1
#if XCHAL_HAVE_LOOPS
loopnez \ar, \endlabelref
#else /* XCHAL_HAVE_LOOPS */
beqz \ar, \endlabelref
\startlabel:
addi \ar, \ar, -1
#endif /* XCHAL_HAVE_LOOPS */
.endm // floopnez_
.macro floopgtz_ ar, startlabel, endlabelref
.ifdef _infloop_
.if _infloop_
.err // Error: floopgtz cannot be nested
.endif
.endif
.set _infloop_, 1
#if XCHAL_HAVE_LOOPS
loopgtz \ar, \endlabelref
#else /* XCHAL_HAVE_LOOPS */
bltz \ar, \endlabelref
beqz \ar, \endlabelref
\startlabel:
addi \ar, \ar, -1
#endif /* XCHAL_HAVE_LOOPS */
.endm // floopgtz_
.macro floopend_ ar, startlabelref, endlabel
.ifndef _infloop_
.err // Error: floopend without matching floopXXX
.endif
.ifeq _infloop_
.err // Error: floopend without matching floopXXX
.endif
.set _infloop_, 0
#if ! XCHAL_HAVE_LOOPS
bnez \ar, \startlabelref
#endif /* XCHAL_HAVE_LOOPS */
\endlabel:
.endm // floopend_
/*----------------------------------------------------------------------
* crsil -- conditional RSIL (read/set interrupt level)
*
* Executes the RSIL instruction if it exists, else just reads PS.
* The RSIL instruction does not exist in the new exception architecture
* if the interrupt option is not selected.
*/
.macro crsil ar, newlevel
#if XCHAL_HAVE_OLD_EXC_ARCH || XCHAL_HAVE_INTERRUPTS
rsil \ar, \newlevel
#else
rsr \ar, PS
#endif
.endm // crsil
/*----------------------------------------------------------------------
* safe_movi_a0 -- move constant into a0 when L32R is not safe
*
* This macro is typically used by interrupt/exception handlers.
* Loads a 32-bit constant in a0, without using any other register,
* and without corrupting the LITBASE register, even when the
* value of the LITBASE register is unknown (eg. when application
* code and interrupt/exception handling code are built independently,
* and thus with independent values of the LITBASE register;
* debug monitors are one example of this).
*
* Worst-case size of resulting code: 17 bytes.
*/
.macro safe_movi_a0 constant
#if XCHAL_HAVE_ABSOLUTE_LITERALS
/* Contort a PC-relative literal load even though we may be in litbase-relative mode: */
j 1f
.begin no-transform // ensure what follows is assembled exactly as-is
.align 4 // ensure constant and call0 target ...
.byte 0 // ... are 4-byte aligned (call0 instruction is 3 bytes long)
1: call0 2f // read PC (that follows call0) in a0
.long \constant // 32-bit constant to load into a0
2:
.end no-transform
l32i a0, a0, 0 // load constant
#else
movi a0, \constant // no LITBASE, can assume PC-relative L32R
#endif
.endm
/*----------------------------------------------------------------------
* window_spill{4,8,12}
*
* These macros spill callers' register windows to the stack.
* They work for both privileged and non-privileged tasks.
* Must be called from a windowed ABI context, eg. within
* a windowed ABI function (ie. valid stack frame, window
* exceptions enabled, not in exception mode, etc).
*
* This macro requires a single invocation of the window_spill_common
* macro in the same assembly unit and section.
*
* Note that using window_spill{4,8,12} macros is more efficient
* than calling a function implemented using window_spill_function,
* because the latter needs extra code to figure out the size of
* the call to the spilling function.
*
* Example usage:
*
* .text
* .align 4
* .global some_function
* .type some_function,@function
* some_function:
* entry a1, 16
* :
* :
*
* window_spill4 // Spill windows of some_function's callers; preserves a0..a3 only;
* // to use window_spill{8,12} in this example function we'd have
* // to increase space allocated by the entry instruction, because
* // 16 bytes only allows call4; 32 or 48 bytes (+locals) are needed
* // for call8/window_spill8 or call12/window_spill12 respectively.
*
* :
*
* retw
*
* window_spill_common // instantiates code used by window_spill4
*
*
* On entry:
* none (if window_spill4)
* stack frame has enough space allocated for call8 (if window_spill8)
* stack frame has enough space allocated for call12 (if window_spill12)
* On exit:
* a4..a15 clobbered (if window_spill4)
* a8..a15 clobbered (if window_spill8)
* a12..a15 clobbered (if window_spill12)
* no caller windows are in live registers
*/
.macro window_spill4
#if XCHAL_HAVE_WINDOWED
# if XCHAL_NUM_AREGS == 16
movi a15, 0 // for 16-register files, no need to call to reach the end
# elif XCHAL_NUM_AREGS == 32
call4 .L__wdwspill_assist28 // call deep enough to clear out any live callers
# elif XCHAL_NUM_AREGS == 64
call4 .L__wdwspill_assist60 // call deep enough to clear out any live callers
# endif
#endif
.endm // window_spill4
.macro window_spill8
#if XCHAL_HAVE_WINDOWED
# if XCHAL_NUM_AREGS == 16
movi a15, 0 // for 16-register files, no need to call to reach the end
# elif XCHAL_NUM_AREGS == 32
call8 .L__wdwspill_assist24 // call deep enough to clear out any live callers
# elif XCHAL_NUM_AREGS == 64
call8 .L__wdwspill_assist56 // call deep enough to clear out any live callers
# endif
#endif
.endm // window_spill8
.macro window_spill12
#if XCHAL_HAVE_WINDOWED
# if XCHAL_NUM_AREGS == 16
movi a15, 0 // for 16-register files, no need to call to reach the end
# elif XCHAL_NUM_AREGS == 32
call12 .L__wdwspill_assist20 // call deep enough to clear out any live callers
# elif XCHAL_NUM_AREGS == 64
call12 .L__wdwspill_assist52 // call deep enough to clear out any live callers
# endif
#endif
.endm // window_spill12
/*----------------------------------------------------------------------
* window_spill_function
*
* This macro outputs a function that will spill its caller's callers'
* register windows to the stack. Eg. it could be used to implement
* a version of xthal_window_spill() that works in non-privileged tasks.
* This works for both privileged and non-privileged tasks.
*
* Typical usage:
*
* .text
* .align 4
* .global my_spill_function
* .type my_spill_function,@function
* my_spill_function:
* window_spill_function
*
* On entry to resulting function:
* none
* On exit from resulting function:
* none (no caller windows are in live registers)
*/
.macro window_spill_function
#if XCHAL_HAVE_WINDOWED
# if XCHAL_NUM_AREGS == 32
entry sp, 48
bbci.l a0, 31, 1f // branch if called with call4
bbsi.l a0, 30, 2f // branch if called with call12
call8 .L__wdwspill_assist16 // called with call8, only need another 8
retw
1: call12 .L__wdwspill_assist16 // called with call4, only need another 12
retw
2: call4 .L__wdwspill_assist16 // called with call12, only need another 4
retw
# elif XCHAL_NUM_AREGS == 64
entry sp, 48
bbci.l a0, 31, 1f // branch if called with call4
bbsi.l a0, 30, 2f // branch if called with call12
call4 .L__wdwspill_assist52 // called with call8, only need a call4
retw
1: call8 .L__wdwspill_assist52 // called with call4, only need a call8
retw
2: call12 .L__wdwspill_assist40 // called with call12, can skip a call12
retw
# elif XCHAL_NUM_AREGS == 16
entry sp, 16
bbci.l a0, 31, 1f // branch if called with call4
bbsi.l a0, 30, 2f // branch if called with call12
movi a7, 0 // called with call8
retw
1: movi a11, 0 // called with call4
2: retw // if called with call12, everything already spilled
// movi a15, 0 // trick to spill all but the direct caller
// j 1f
// // The entry instruction is magical in the assembler (gets auto-aligned)
// // so we have to jump to it to avoid falling through the padding.
// // We need entry/retw to know where to return.
//1: entry sp, 16
// retw
# else
# error "unrecognized address register file size"
# endif
#endif /* XCHAL_HAVE_WINDOWED */
window_spill_common
.endm // window_spill_function
/*----------------------------------------------------------------------
* window_spill_common
*
* Common code used by any number of invocations of the window_spill##
* and window_spill_function macros.
*
* Must be instantiated exactly once within a given assembly unit,
* within call/j range of and same section as window_spill##
* macro invocations for that assembly unit.
* (Is automatically instantiated by the window_spill_function macro.)
*/
.macro window_spill_common
#if XCHAL_HAVE_WINDOWED && (XCHAL_NUM_AREGS == 32 || XCHAL_NUM_AREGS == 64)
.ifndef .L__wdwspill_defined
# if XCHAL_NUM_AREGS >= 64
.L__wdwspill_assist60:
entry sp, 32
call8 .L__wdwspill_assist52
retw
.L__wdwspill_assist56:
entry sp, 16
call4 .L__wdwspill_assist52
retw
.L__wdwspill_assist52:
entry sp, 48
call12 .L__wdwspill_assist40
retw
.L__wdwspill_assist40:
entry sp, 48
call12 .L__wdwspill_assist28
retw
# endif
.L__wdwspill_assist28:
entry sp, 48
call12 .L__wdwspill_assist16
retw
.L__wdwspill_assist24:
entry sp, 32
call8 .L__wdwspill_assist16
retw
.L__wdwspill_assist20:
entry sp, 16
call4 .L__wdwspill_assist16
retw
.L__wdwspill_assist16:
entry sp, 16
movi a15, 0
retw
.set .L__wdwspill_defined, 1
.endif
#endif /* XCHAL_HAVE_WINDOWED with 32 or 64 aregs */
.endm // window_spill_common
/*----------------------------------------------------------------------
* beqi32
*
* macro implements version of beqi for arbitrary 32-bit immediate value
*
* beqi32 ax, ay, imm32, label
*
* Compares value in register ax with imm32 value and jumps to label if
* equal. Clobbers register ay if needed
*
*/
.macro beqi32 ax, ay, imm, label
.ifeq ((\imm-1) & ~7) // 1..8 ?
beqi \ax, \imm, \label
.else
.ifeq (\imm+1) // -1 ?
beqi \ax, \imm, \label
.else
.ifeq (\imm) // 0 ?
beqz \ax, \label
.else
// We could also handle immediates 10,12,16,32,64,128,256
// but it would be a long macro...
movi \ay, \imm
beq \ax, \ay, \label
.endif
.endif
.endif
.endm // beqi32
/*----------------------------------------------------------------------
* isync_retw_nop
*
* This macro must be invoked immediately after ISYNC if ISYNC
* would otherwise be immediately followed by RETW (or other instruction
* modifying WindowBase or WindowStart), in a context where
* kernel vector mode may be selected, and level-one interrupts
* and window overflows may be enabled, on an XEA1 configuration.
*
* On hardware with erratum "XEA1KWIN" (see <xtensa/core.h> for details),
* XEA1 code must have at least one instruction between ISYNC and RETW if
* run in kernel vector mode with interrupts and window overflows enabled.
*/
.macro isync_retw_nop
#if XCHAL_MAYHAVE_ERRATUM_XEA1KWIN
nop
#endif
.endm
/*----------------------------------------------------------------------
* isync_erratum453
*
* This macro must be invoked at certain points in the code,
* such as in exception and interrupt vectors in particular,
* to work around erratum 453.
*/
.macro isync_erratum453
#if XCHAL_ERRATUM_453
isync
#endif
.endm
/*----------------------------------------------------------------------
* abs
*
* implements abs on machines that do not have it configured
*/
#if !XCHAL_HAVE_ABS
.macro abs arr, ars
.ifc \arr, \ars
//src equal dest is less efficient
bgez \arr, 1f
neg \arr, \arr
1:
.else
neg \arr, \ars
movgez \arr, \ars, \ars
.endif
.endm
#endif /* !XCHAL_HAVE_ABS */
/*----------------------------------------------------------------------
* addx2
*
* implements addx2 on machines that do not have it configured
*
*/
#if !XCHAL_HAVE_ADDX
.macro addx2 arr, ars, art
.ifc \arr, \art
.ifc \arr, \ars
// addx2 a, a, a (not common)
.err
.else
add \arr, \ars, \art
add \arr, \ars, \art
.endif
.else
//addx2 a, b, c
//addx2 a, a, b
//addx2 a, b, b
slli \arr, \ars, 1
add \arr, \arr, \art
.endif
.endm
#endif /* !XCHAL_HAVE_ADDX */
/*----------------------------------------------------------------------
* addx4
*
* implements addx4 on machines that do not have it configured
*
*/
#if !XCHAL_HAVE_ADDX
.macro addx4 arr, ars, art
.ifc \arr, \art
.ifc \arr, \ars
// addx4 a, a, a (not common)
.err
.else
//# addx4 a, b, a
add \arr, \ars, \art
add \arr, \ars, \art
add \arr, \ars, \art
add \arr, \ars, \art
.endif
.else
//addx4 a, b, c
//addx4 a, a, b
//addx4 a, b, b
slli \arr, \ars, 2
add \arr, \arr, \art
.endif
.endm
#endif /* !XCHAL_HAVE_ADDX */
/*----------------------------------------------------------------------
* addx8
*
* implements addx8 on machines that do not have it configured
*
*/
#if !XCHAL_HAVE_ADDX
.macro addx8 arr, ars, art
.ifc \arr, \art
.ifc \arr, \ars
//addx8 a, a, a (not common)
.err
.else
//addx8 a, b, a
add \arr, \ars, \art
add \arr, \ars, \art
add \arr, \ars, \art
add \arr, \ars, \art
add \arr, \ars, \art
add \arr, \ars, \art
add \arr, \ars, \art
add \arr, \ars, \art
.endif
.else
//addx8 a, b, c
//addx8 a, a, b
//addx8 a, b, b
slli \arr, \ars, 3
add \arr, \arr, \art
.endif
.endm
#endif /* !XCHAL_HAVE_ADDX */
/*----------------------------------------------------------------------
* rfe_rfue
*
* Maps to RFUE on XEA1, and RFE on XEA2. No mapping on XEAX.
*/
#if XCHAL_HAVE_XEA1
.macro rfe_rfue
rfue
.endm
#elif XCHAL_HAVE_XEA2
.macro rfe_rfue
rfe
.endm
#endif
/*----------------------------------------------------------------------
* abi_entry
*
* Generate proper function entry sequence for the current ABI
* (windowed or call0). Takes care of allocating stack space (up to 1kB)
* and saving the return PC, if necessary. The corresponding abi_return
* macro does the corresponding stack deallocation and restoring return PC.
*
* Parameters are:
*
* locsize Number of bytes to allocate on the stack
* for local variables (and for args to pass to
* callees, if any calls are made). Defaults to zero.
* The macro rounds this up to a multiple of 16.
* NOTE: large values are allowed (e.g. up to 1 GB).
*
* callsize Maximum call size made by this function.
* Leave zero (default) for leaf functions, i.e. if
* this function makes no calls to other functions.
* Otherwise must be set to 4, 8, or 12 according
* to whether the "largest" call made is a call[x]4,
* call[x]8, or call[x]12 (for call0 ABI, it makes
* no difference whether this is set to 4, 8 or 12,
* but it must be set to one of these values).
*
* NOTE: It is up to the caller to align the entry point, declare the
* function symbol, make it global, etc.
*
* NOTE: This macro relies on assembler relaxation for large values
* of locsize. It might not work with the no-transform directive.
* NOTE: For the call0 ABI, this macro ensures SP is allocated or
* de-allocated cleanly, i.e. without temporarily allocating too much
* (or allocating negatively!) due to addi relaxation.
*
* NOTE: Generating the proper sequence and register allocation for
* making calls in an ABI independent manner is a separate topic not
* covered by this macro.
*
* NOTE: To access arguments, you can't use a fixed offset from SP.
* The offset depends on the ABI, whether the function is leaf, etc.
* The simplest method is probably to use the .locsz symbol, which
* is set by this macro to the actual number of bytes allocated on
* the stack, in other words, to the offset from SP to the arguments.
* E.g. for a function whose arguments are all 32-bit integers, you
* can get the 7th and 8th arguments (1st and 2nd args stored on stack)
* using:
* l32i a2, sp, .locsz
* l32i a3, sp, .locsz+4
* (this example works as long as locsize is under L32I's offset limit
* of 1020 minus up to 48 bytes of ABI-specific stack usage;
* otherwise you might first need to do "addi a?, sp, .locsz"
* or similar sequence).
*
* NOTE: For call0 ABI, this macro (and abi_return) may clobber a9
* (a caller-saved register).
*
* Examples:
* abi_entry
* abi_entry 5
* abi_entry 22, 8
* abi_entry 0, 4
*/
/*
* Compute .locsz and .callsz without emitting any instructions.
* Used by both abi_entry and abi_return.
* Assumes locsize >= 0.
*/
.macro abi_entry_size locsize=0, callsize=0
#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
.ifeq \callsize
.set .callsz, 16
.else
.ifeq \callsize-4
.set .callsz, 16
.else
.ifeq \callsize-8
.set .callsz, 32
.else
.ifeq \callsize-12
.set .callsz, 48
.else
.error "abi_entry: invalid call size \callsize"
.endif
.endif
.endif
.endif
.set .locsz, .callsz + ((\locsize + 15) & -16)
#else
.set .callsz, \callsize
.if .callsz /* if calls, need space for return PC */
.set .locsz, (\locsize + 4 + 15) & -16
.else
.set .locsz, (\locsize + 15) & -16
.endif
#endif
.endm
.macro abi_entry locsize=0, callsize=0
.iflt \locsize
.error "abi_entry: invalid negative size of locals (\locsize)"
.endif
abi_entry_size \locsize, \callsize
#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
.ifgt .locsz - 32760 /* .locsz > 32760 (ENTRY's max range)? */
/* Funky computation to try to have assembler use addmi efficiently if possible: */
entry sp, 0x7F00 + (.locsz & 0xF0)
addi a12, sp, - ((.locsz & -0x100) - 0x7F00)
movsp sp, a12
.else
entry sp, .locsz
.endif
#else
.if .locsz
.ifle .locsz - 128 /* if locsz <= 128 */
addi sp, sp, -.locsz
.if .callsz
s32i a0, sp, .locsz - 4
.endif
.elseif .callsz /* locsz > 128, with calls: */
movi a9, .locsz - 16 /* note: a9 is caller-saved */
addi sp, sp, -16
s32i a0, sp, 12
sub sp, sp, a9
.else /* locsz > 128, no calls: */
movi a9, .locsz
sub sp, sp, a9
.endif /* end */
.endif
#endif
.endm
/*----------------------------------------------------------------------
* abi_return
*
* Generate proper function exit sequence for the current ABI
* (windowed or call0). Takes care of freeing stack space and
* restoring the return PC, if necessary.
* NOTE: This macro MUST be invoked following a corresponding
* abi_entry macro invocation. For call0 ABI in particular,
* all stack and PC restoration are done according to the last
* abi_entry macro invoked before this macro in the assembly file.
*
* Normally this macro takes no arguments. However to allow
* for placing abi_return *before* abi_entry (as must be done
* for some highly optimized assembly), it optionally takes
* exactly the same arguments as abi_entry.
*/
.macro abi_return locsize=-1, callsize=0
.ifge \locsize
abi_entry_size \locsize, \callsize
.endif
#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
retw
#else
.if .locsz
.iflt .locsz - 128 /* if locsz < 128 */
.if .callsz
l32i a0, sp, .locsz - 4
.endif
addi sp, sp, .locsz
.elseif .callsz /* locsz >= 128, with calls: */
addi a9, sp, .locsz - 16
l32i a0, a9, 12
addi sp, a9, 16
.else /* locsz >= 128, no calls: */
movi a9, .locsz
add sp, sp, a9
.endif /* end */
.endif
ret
#endif
.endm
/*
* HW erratum fixes.
*/
.macro hw_erratum_487_fix
#if defined XSHAL_ERRATUM_487_FIX
isync
#endif
.endm
#endif /*XTENSA_COREASM_H*/