xnu-792.13.8.tar.gz

[apple/xnu.git] / osfmk / i386 / commpage / bcopy_sse3.s

/*
 * Copyright (c) 2005 Apple Computer, Inc. All rights reserved.
 *
 * @APPLE_LICENSE_OSREFERENCE_HEADER_START@
 * 
 * This file contains Original Code and/or Modifications of Original Code 
 * as defined in and that are subject to the Apple Public Source License 
 * Version 2.0 (the 'License'). You may not use this file except in 
 * compliance with the License.  The rights granted to you under the 
 * License may not be used to create, or enable the creation or 
 * redistribution of, unlawful or unlicensed copies of an Apple operating 
 * system, or to circumvent, violate, or enable the circumvention or 
 * violation of, any terms of an Apple operating system software license 
 * agreement.
 *
 * Please obtain a copy of the License at 
 * http://www.opensource.apple.com/apsl/ and read it before using this 
 * file.
 *
 * The Original Code and all software distributed under the License are 
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 
 * Please see the License for the specific language governing rights and 
 * limitations under the License.
 *
 * @APPLE_LICENSE_OSREFERENCE_HEADER_END@
 */
        
#include <machine/cpu_capabilities.h>
#include <machine/commpage.h>

/*
 * The bcopy/memcpy loops, tuned for Pentium-M class processors with SSE3
 * and 64-byte cache lines, such as Core and Core 2.
 *
 * The following #defines are tightly coupled to the u-architecture:
 */

#define kShort  80			// too short to bother with SSE (must be >=80)
#define kVeryLong   (500*1024)          // large enough for non-temporal stores (must be >= 8192)
#define kBigChunk   (256*1024)          // outer loop chunk size for kVeryLong sized operands
#define kFastUCode  (16*1024)		// cutoff for microcode fastpath for "rep/movsl"


// void bcopy(const void *src, void *dst, size_t len);
 
        .text
        .align 5, 0x90
Lbcopy_sse3:				// void bcopy(const void *src, void *dst, size_t len)
	pushl	%ebp			// set up a frame for backtraces
	movl	%esp,%ebp
        pushl   %esi
        pushl   %edi
        movl    8(%ebp),%esi		// get source ptr
        movl    12(%ebp),%edi           // get dest ptr
        jmp     Ljoin 

//
// void *memcpy(void *dst, const void *src, size_t len);
// void *memmove(void *dst, const void *src, size_t len);
//
// NB: These need to be 32 bytes from bcopy():
//

        .align	5, 0x90
Lmemcpy:				// void *memcpy(void *dst, const void *src, size_t len)
Lmemmove:				// void *memmove(void *dst, const void *src, size_t len)
	pushl	%ebp			// set up a frame for backtraces
	movl	%esp,%ebp
        pushl   %esi
        pushl   %edi
        movl    8(%ebp),%edi		// get dest ptr
        movl    12(%ebp),%esi           // get source ptr
        
Ljoin:                                  // here from bcopy() with esi and edi loaded
        movl    16(%ebp),%ecx           // get length
        movl    %edi,%edx
        subl    %esi,%edx               // (dest - source)
        cmpl    %ecx,%edx               // must move in reverse if (dest - source) < length
        jb      LReverseIsland
Lrejoin:                                // here from very-long-operand copies
        cmpl    $(kShort),%ecx          // long enough to bother with SSE?
        ja      LNotShort               // yes
        
// Handle short forward copies.  As the most common case, this is the fall-through path.
//      ecx = length (<= kShort)
//      esi = source ptr
//      edi = dest ptr

Lshort:
	movl    %ecx,%edx		// copy length
	shrl	$2,%ecx			// get #doublewords
	jz	LLeftovers
2:					// loop copying doublewords
	movl	(%esi),%eax
	addl	$4,%esi
	movl	%eax,(%edi)
	addl	$4,%edi
	dec	%ecx
	jnz	2b
LLeftovers:				// handle leftover bytes (0..3) in last word
	andl	$3,%edx			// any leftover bytes?
	jz	5f
4:					// loop copying bytes
	movb	(%esi),%al
	inc	%esi
	movb	%al,(%edi)
	inc	%edi
	dec	%edx
	jnz	4b
5:
        movl    8(%ebp),%eax		// get return value (dst ptr) for memcpy/memmove
        popl    %edi
        popl    %esi
	popl	%ebp
        ret


LReverseIsland:				// keep the "jb" above a short branch...
	jmp	LReverse		// ...because reverse moves are uncommon


// Handle forward moves that are long enough to justify use of SSE3.
// First, 16-byte align the destination.
//      ecx = length (> kShort)
//      esi = source ptr
//      edi = dest ptr

LNotShort:
        cmpl    $(kVeryLong),%ecx       // long enough to justify heavyweight loops?
        movl    %edi,%edx               // copy destination
        jae     LVeryLong		// use very-long-operand path
        negl    %edx
        andl    $15,%edx                // get #bytes to align destination
	jz	LDestAligned		// already aligned
        subl    %edx,%ecx               // decrement length
1:					// loop copying 1..15 bytes
	movb	(%esi),%al
	inc	%esi
	movb	%al,(%edi)
	inc	%edi
	dec	%edx
	jnz	1b
        
// Destination is now aligned.  Prepare for forward loops over 64-byte chunks.
// Since kShort>=80 and we've moved at most 15 bytes already, there is at least one chunk.

LDestAligned:
        movl    %ecx,%edx               // copy length
	movl	%ecx,%eax		// twice
        andl    $63,%ecx                // get remaining bytes for Lshort
        andl    $-64,%edx               // get number of bytes we will copy in inner loop
        addl    %edx,%esi               // point to 1st byte not copied
        addl    %edx,%edi
        negl    %edx                    // now generate offset to 1st byte to be copied
	testl	$15,%esi		// is source aligned too?
	jnz	LUnalignedLoop		// no


	cmpl	$(kFastUCode),%eax	// long enough for the fastpath in microcode?
	jb	LAlignedLoop		// no, use SSE
	cld				// we'll move forward
	movl	%eax,%ecx		// copy length again
	shrl	$2,%ecx			// compute #words to move
        addl    %edx,%esi               // restore ptrs to 1st byte of source and dest
        addl    %edx,%edi
	rep				// the u-code will optimize this
	movsl
	movl	%eax,%edx		// original length
	jmp	LLeftovers		// handle 0..3 leftover bytes


// Forward aligned loop for medium length operands (kShort < n < kVeryLong).

	.align	4,0x90			// 16-byte align inner loops
LAlignedLoop:                           // loop over 64-byte chunks
        movdqa  (%esi,%edx),%xmm0
        movdqa  16(%esi,%edx),%xmm1
        movdqa  32(%esi,%edx),%xmm2
        movdqa  48(%esi,%edx),%xmm3

        movdqa  %xmm0,(%edi,%edx)
        movdqa  %xmm1,16(%edi,%edx)
        movdqa  %xmm2,32(%edi,%edx)
        movdqa  %xmm3,48(%edi,%edx)
        
        addl    $64,%edx
        jnz     LAlignedLoop
        
        jmp     Lshort                  // copy remaining 0..15 bytes and done
            

// Forward unaligned loop for medium length operands (kShort < n < kVeryLong).
// Note that LDDQU==MOVDQU on these machines, ie we don't care when we cross
// source cache lines.

	.align	4,0x90			// 16-byte align inner loops
LUnalignedLoop:                         // loop over 64-byte chunks
        movdqu  (%esi,%edx),%xmm0	// the loads are unaligned
        movdqu  16(%esi,%edx),%xmm1
        movdqu  32(%esi,%edx),%xmm2
        movdqu  48(%esi,%edx),%xmm3
        
        movdqa  %xmm0,(%edi,%edx)	// we can use aligned stores
        movdqa  %xmm1,16(%edi,%edx)
        movdqa  %xmm2,32(%edi,%edx)
        movdqa  %xmm3,48(%edi,%edx)
        
        addl    $64,%edx
        jnz     LUnalignedLoop
        
        jmp     Lshort                  // copy remaining 0..63 bytes and done
        

// Very long forward moves.  These are at least several pages, so we loop over big
// chunks of memory (kBigChunk in size.)  We first prefetch the chunk, and then copy
// it using non-temporal stores.  Hopefully all the reads occur in the prefetch loop,
// so the copy loop reads from L2 and writes directly to memory (with write combining.)
// This minimizes bus turnaround and maintains good DRAM page locality.
// Note that for this scheme to work, kVeryLong must be a large fraction of L2 cache
// size.  Otherwise, it is counter-productive to bypass L2 on the stores.
//      ecx = length (>= kVeryLong bytes)
//      edi = dest (aligned)
//      esi = source

LVeryLong:
	pushl	%ebx			// we'll need to use this
	movl	%edi,%ebx		// copy dest ptr
	negl	%ebx
	andl	$63,%ebx		// get #bytes to cache line align destination
	jz	LBigChunkLoop		// already aligned
	
// Cache line align destination, so temporal stores in copy loops work right.

	pushl	%ecx			// save total length remaining
	pushl	%ebx			// arg3 - #bytes to align destination (1..63)
	pushl	%esi			// arg2 - source
	pushl	%edi			// arg1 - dest
	call	Lmemcpy			// align the destination
	movl	12(%esp),%ecx		// recover total length
	addl	$16,%esp
	addl	%ebx,%esi		// adjust ptrs and lengths past copy
	addl	%ebx,%edi
	subl	%ebx,%ecx
	
// Loop over big chunks.
//      ecx = length remaining (>= 4096)
//      edi = dest (64-byte aligned)
//      esi = source (may be unaligned)
	
LBigChunkLoop:
        movl    $(kBigChunk),%edx       // assume we can do a full chunk
	cmpl	%edx,%ecx		// do we have a full chunk left to do?
	cmovbl	%ecx,%edx		// if not, only move what we have left
	andl	$-4096,%edx		// we work in page multiples
	xor	%eax,%eax		// initialize chunk offset
	jmp	LTouchLoop

// Because the source may be unaligned, we use byte loads to touch.
//      ecx = length remaining (including this chunk)
//      edi = ptr to start of dest chunk
//      esi = ptr to start of source chunk
//	edx = chunk length (multiples of pages)
//	ebx = scratch reg used to read a byte of each cache line
//	eax = chunk offset

	.align	4,0x90			// 16-byte align inner loops
LTouchLoop:
	movzb	(%esi,%eax),%ebx	// touch line 0, 2, 4, or 6 of page
	movzb	1*64(%esi,%eax),%ebx	// touch line 1, 3, 5, or 7
	movzb	8*64(%esi,%eax),%ebx	// touch line 8, 10, 12, or 14
	movzb	9*64(%esi,%eax),%ebx	// etc
	
	movzb	16*64(%esi,%eax),%ebx
	movzb	17*64(%esi,%eax),%ebx
	movzb	24*64(%esi,%eax),%ebx
	movzb	25*64(%esi,%eax),%ebx

	movzb	32*64(%esi,%eax),%ebx
	movzb	33*64(%esi,%eax),%ebx
	movzb	40*64(%esi,%eax),%ebx
	movzb	41*64(%esi,%eax),%ebx

	movzb	48*64(%esi,%eax),%ebx
	movzb	49*64(%esi,%eax),%ebx
	movzb	56*64(%esi,%eax),%ebx
	movzb	57*64(%esi,%eax),%ebx
	
	subl	$-128,%eax		// next slice of page (adding 128 w 8-bit immediate)
	testl	$512,%eax		// done with this page?
	jz	LTouchLoop		// no, next of four slices
	addl	$(4096-512),%eax	// move on to next page
	cmpl	%eax,%edx		// done with this chunk?
	jnz	LTouchLoop		// no, do next page
	
// The chunk has been pre-fetched, now copy it using non-temporal stores.
// There are two copy loops, depending on whether the source is 16-byte aligned
// or not.

	addl	%edx,%esi		// increment ptrs by chunk length
	addl	%edx,%edi
	subl	%edx,%ecx		// adjust remaining length
	negl	%edx			// prepare loop index (counts up to 0)
	testl	$15,%esi		// is source 16-byte aligned?
	jnz	LVeryLongUnaligned	// source is not aligned
	jmp	LVeryLongAligned

	.align	4,0x90			// 16-byte align inner loops
LVeryLongAligned:			// aligned loop over 128-bytes
        movdqa  (%esi,%edx),%xmm0
        movdqa  16(%esi,%edx),%xmm1
        movdqa  32(%esi,%edx),%xmm2
        movdqa  48(%esi,%edx),%xmm3
        movdqa  64(%esi,%edx),%xmm4
        movdqa  80(%esi,%edx),%xmm5
        movdqa  96(%esi,%edx),%xmm6
        movdqa  112(%esi,%edx),%xmm7

        movntdq %xmm0,(%edi,%edx)
        movntdq %xmm1,16(%edi,%edx)
        movntdq %xmm2,32(%edi,%edx)
        movntdq %xmm3,48(%edi,%edx)
        movntdq %xmm4,64(%edi,%edx)
        movntdq %xmm5,80(%edi,%edx)
        movntdq %xmm6,96(%edi,%edx)
        movntdq %xmm7,112(%edi,%edx)
        
        subl    $-128,%edx		// add 128 with an 8-bit immediate
        jnz	LVeryLongAligned
	jmp	LVeryLongChunkEnd

	.align	4,0x90			// 16-byte align inner loops
LVeryLongUnaligned:			// unaligned loop over 128-bytes
        movdqu  (%esi,%edx),%xmm0
        movdqu  16(%esi,%edx),%xmm1
        movdqu  32(%esi,%edx),%xmm2
        movdqu  48(%esi,%edx),%xmm3
        movdqu  64(%esi,%edx),%xmm4
        movdqu  80(%esi,%edx),%xmm5
        movdqu  96(%esi,%edx),%xmm6
        movdqu  112(%esi,%edx),%xmm7

        movntdq %xmm0,(%edi,%edx)
        movntdq %xmm1,16(%edi,%edx)
        movntdq %xmm2,32(%edi,%edx)
        movntdq %xmm3,48(%edi,%edx)
        movntdq %xmm4,64(%edi,%edx)
        movntdq %xmm5,80(%edi,%edx)
        movntdq %xmm6,96(%edi,%edx)
        movntdq %xmm7,112(%edi,%edx)
        
        subl    $-128,%edx		// add 128 with an 8-bit immediate
        jnz     LVeryLongUnaligned

LVeryLongChunkEnd:
	cmpl	$4096,%ecx		// at least another page to go?
	jae	LBigChunkLoop		// yes
	
	sfence				// required by non-temporal stores
	popl	%ebx
        jmp     Lrejoin                 // handle remaining (0..4095) bytes


// Reverse moves.
//      ecx = length
//      esi = source ptr
//      edi = dest ptr

LReverse:
        addl    %ecx,%esi               // point to end of strings
        addl    %ecx,%edi
        cmpl    $(kShort),%ecx          // long enough to bother with SSE?
        ja      LReverseNotShort        // yes

// Handle reverse short copies.
//      ecx = length
//      esi = one byte past end of source
//      edi = one byte past end of dest

LReverseShort:
	movl    %ecx,%edx		// copy length
	shrl	$2,%ecx			// #words
	jz	3f
1:
	subl	$4,%esi
	movl	(%esi),%eax
	subl	$4,%edi
	movl	%eax,(%edi)
	dec	%ecx
	jnz	1b
3:
	andl	$3,%edx			// bytes?
	jz	5f
4:
	dec	%esi
	movb	(%esi),%al
	dec	%edi
	movb	%al,(%edi)
	dec	%edx
	jnz	4b
5:
        movl    8(%ebp),%eax		// get return value (dst ptr) for memcpy/memmove
        popl    %edi
        popl    %esi
	popl	%ebp
        ret

// Handle a reverse move long enough to justify using SSE.
//      ecx = length
//      esi = one byte past end of source
//      edi = one byte past end of dest

LReverseNotShort:
        movl    %edi,%edx               // copy destination
        andl    $15,%edx                // get #bytes to align destination
        je      LReverseDestAligned     // already aligned
        subl	%edx,%ecx		// adjust length
1:					// loop copying 1..15 bytes
	dec	%esi
	movb	(%esi),%al
	dec	%edi
	movb	%al,(%edi)
	dec	%edx
	jnz	1b
        
// Destination is now aligned.  Prepare for reverse loops.

LReverseDestAligned:
        movl    %ecx,%edx               // copy length
        andl    $63,%ecx                // get remaining bytes for Lshort
        andl    $-64,%edx               // get number of bytes we will copy in inner loop
        subl    %edx,%esi               // point to endpoint of copy
        subl    %edx,%edi
	testl	$15,%esi		// is source aligned too?
        jnz     LReverseUnalignedLoop   // no
	jmp	LReverseAlignedLoop	// use aligned loop

	.align	4,0x90			// 16-byte align inner loops
LReverseAlignedLoop:                    // loop over 64-byte chunks
        movdqa  -16(%esi,%edx),%xmm0
        movdqa  -32(%esi,%edx),%xmm1
        movdqa  -48(%esi,%edx),%xmm2
        movdqa  -64(%esi,%edx),%xmm3

        movdqa  %xmm0,-16(%edi,%edx)
        movdqa  %xmm1,-32(%edi,%edx)
        movdqa  %xmm2,-48(%edi,%edx)
        movdqa  %xmm3,-64(%edi,%edx)
        
        subl    $64,%edx
        jne     LReverseAlignedLoop
        
        jmp     LReverseShort           // copy remaining 0..63 bytes and done

    
// Reverse, unaligned loop.  LDDQU==MOVDQU on these machines.
        
	.align	4,0x90			// 16-byte align inner loops
LReverseUnalignedLoop:                  // loop over 64-byte chunks
        movdqu  -16(%esi,%edx),%xmm0
        movdqu  -32(%esi,%edx),%xmm1
        movdqu  -48(%esi,%edx),%xmm2
        movdqu  -64(%esi,%edx),%xmm3
        
        movdqa  %xmm0,-16(%edi,%edx)
        movdqa  %xmm1,-32(%edi,%edx)
        movdqa  %xmm2,-48(%edi,%edx)
        movdqa  %xmm3,-64(%edi,%edx)
        
        subl    $64,%edx
        jne     LReverseUnalignedLoop
        
        jmp     LReverseShort           // copy remaining 0..63 bytes and done


	COMMPAGE_DESCRIPTOR(bcopy_sse3,_COMM_PAGE_BCOPY,kHasSSE2+kCache64,kHasSupplementalSSE3)