[mirror_edk2.git] / MdePkg / Library / BaseMemoryLibOptDxe / Arm / ScanMem.S

// Copyright (c) 2010-2011, Linaro Limited\r
// All rights reserved.\r
//\r
// Redistribution and use in source and binary forms, with or without\r
// modification, are permitted provided that the following conditions\r
// are met:\r
//\r
//    * Redistributions of source code must retain the above copyright\r
//    notice, this list of conditions and the following disclaimer.\r
//\r
//    * Redistributions in binary form must reproduce the above copyright\r
//    notice, this list of conditions and the following disclaimer in the\r
//    documentation and/or other materials provided with the distribution.\r
//\r
//    * Neither the name of Linaro Limited nor the names of its\r
//    contributors may be used to endorse or promote products derived\r
//    from this software without specific prior written permission.\r
//\r
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS\r
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT\r
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR\r
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT\r
// HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,\r
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT\r
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,\r
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY\r
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\r
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\r
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\r
//\r
\r
//\r
// Written by Dave Gilbert <david.gilbert@linaro.org>\r
//\r
// This memchr routine is optimised on a Cortex-A9 and should work on\r
// all ARMv7 processors.   It has a fast past for short sizes, and has\r
// an optimised path for large data sets; the worst case is finding the\r
// match early in a large data set.\r
//\r
\r
\r
// 2011-02-07 david.gilbert@linaro.org\r
//    Extracted from local git a5b438d861\r
// 2011-07-14 david.gilbert@linaro.org\r
//    Import endianness fix from local git ea786f1b\r
// 2011-12-07 david.gilbert@linaro.org\r
//    Removed unneeded cbz from align loop\r
\r
// this lets us check a flag in a 00/ff byte easily in either endianness\r
#define CHARTSTMASK(c) 1<<(c*8)\r
\r
    .text\r
    .thumb\r
    .syntax unified\r
\r
    .type ASM_PFX(InternalMemScanMem8), %function\r
ASM_GLOBAL ASM_PFX(InternalMemScanMem8)\r
ASM_PFX(InternalMemScanMem8):\r
    // r0 = start of memory to scan\r
    // r1 = length\r
    // r2 = character to look for\r
    // returns r0 = pointer to character or NULL if not found\r
    uxtb    r2, r2        // Don't think we can trust the caller to actually pass a char\r
\r
    cmp     r1, #16       // If it's short don't bother with anything clever\r
    blt     20f\r
\r
    tst     r0, #7        // If it's already aligned skip the next bit\r
    beq     10f\r
\r
    // Work up to an aligned point\r
5:\r
    ldrb    r3, [r0],#1\r
    subs    r1, r1, #1\r
    cmp     r3, r2\r
    beq     50f           // If it matches exit found\r
    tst     r0, #7\r
    bne     5b            // If not aligned yet then do next byte\r
\r
10:\r
    // At this point, we are aligned, we know we have at least 8 bytes to work with\r
    push    {r4-r7}\r
    orr     r2, r2, r2, lsl #8  // expand the match word across to all bytes\r
    orr     r2, r2, r2, lsl #16\r
    bic     r4, r1, #7    // Number of double words to work with\r
    mvns    r7, #0        // all F's\r
    movs    r3, #0\r
\r
15:\r
    ldmia   r0!, {r5,r6}\r
    subs    r4, r4, #8\r
    eor     r5, r5, r2    // Get it so that r5,r6 have 00's where the bytes match the target\r
    eor     r6, r6, r2\r
    uadd8   r5, r5, r7    // Parallel add 0xff - sets the GE bits for anything that wasn't 0\r
    sel     r5, r3, r7    // bytes are 00 for none-00 bytes, or ff for 00 bytes - NOTE INVERSION\r
    uadd8   r6, r6, r7    // Parallel add 0xff - sets the GE bits for anything that wasn't 0\r
    sel     r6, r5, r7    // chained....bytes are 00 for none-00 bytes, or ff for 00 bytes - NOTE INVERSION\r
    cbnz    r6, 60f\r
    bne     15b           // (Flags from the subs above) If not run out of bytes then go around again\r
\r
    pop     {r4-r7}\r
    and     r2, r2, #0xff // Get r2 back to a single character from the expansion above\r
    and     r1, r1, #7    // Leave the count remaining as the number after the double words have been done\r
\r
20:\r
    cbz     r1, 40f       // 0 length or hit the end already then not found\r
\r
21: // Post aligned section, or just a short call\r
    ldrb    r3, [r0], #1\r
    subs    r1, r1, #1\r
    eor     r3, r3, r2    // r3 = 0 if match - doesn't break flags from sub\r
    cbz     r3, 50f\r
    bne     21b           // on r1 flags\r
\r
40:\r
    movs    r0, #0        // not found\r
    bx      lr\r
\r
50:\r
    subs    r0, r0, #1    // found\r
    bx      lr\r
\r
60: // We're here because the fast path found a hit - now we have to track down exactly which word it was\r
    // r0 points to the start of the double word after the one that was tested\r
    // r5 has the 00/ff pattern for the first word, r6 has the chained value\r
    cmp     r5, #0\r
    itte    eq\r
    moveq   r5, r6        // the end is in the 2nd word\r
    subeq   r0, r0, #3    // Points to 2nd byte of 2nd word\r
    subne   r0, r0, #7    // or 2nd byte of 1st word\r
\r
    // r0 currently points to the 3rd byte of the word containing the hit\r
    tst     r5, #CHARTSTMASK(0)     // 1st character\r
    bne     61f\r
    adds    r0, r0, #1\r
    tst     r5, #CHARTSTMASK(1)     // 2nd character\r
    bne     61f\r
    adds    r0, r0 ,#1\r
    tst     r5, #(3 << 15)          // 2nd & 3rd character\r
    // If not the 3rd must be the last one\r
    it      eq\r
    addeq.n r0, r0, #1\r
\r
61:\r
    pop     {r4-r7}\r
    subs    r0, r0, #1\r
    bx      lr\r
Commit	Line	Data
a37f6605 AB	1	// Copyright (c) 2010-2011, Linaro Limited\r
	2	// All rights reserved.\r
	3	//\r
	4	// Redistribution and use in source and binary forms, with or without\r
	5	// modification, are permitted provided that the following conditions\r
	6	// are met:\r
	7	//\r
	8	// * Redistributions of source code must retain the above copyright\r
	9	// notice, this list of conditions and the following disclaimer.\r
	10	//\r
	11	// * Redistributions in binary form must reproduce the above copyright\r
	12	// notice, this list of conditions and the following disclaimer in the\r
	13	// documentation and/or other materials provided with the distribution.\r
	14	//\r
	15	// * Neither the name of Linaro Limited nor the names of its\r
	16	// contributors may be used to endorse or promote products derived\r
	17	// from this software without specific prior written permission.\r
	18	//\r
	19	// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS\r
	20	// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT\r
	21	// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR\r
	22	// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT\r
	23	// HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,\r
	24	// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT\r
	25	// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,\r
	26	// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY\r
	27	// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\r
	28	// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\r
	29	// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\r
	30	//\r
	31	\r
	32	//\r
	33	// Written by Dave Gilbert <david.gilbert@linaro.org>\r
	34	//\r
	35	// This memchr routine is optimised on a Cortex-A9 and should work on\r
	36	// all ARMv7 processors. It has a fast past for short sizes, and has\r
	37	// an optimised path for large data sets; the worst case is finding the\r
	38	// match early in a large data set.\r
	39	//\r
	40	\r
	41	\r
	42	// 2011-02-07 david.gilbert@linaro.org\r
	43	// Extracted from local git a5b438d861\r
	44	// 2011-07-14 david.gilbert@linaro.org\r
	45	// Import endianness fix from local git ea786f1b\r
	46	// 2011-12-07 david.gilbert@linaro.org\r
	47	// Removed unneeded cbz from align loop\r
	48	\r
	49	// this lets us check a flag in a 00/ff byte easily in either endianness\r
	50	#define CHARTSTMASK(c) 1<<(c*8)\r
	51	\r
	52	.text\r
	53	.thumb\r
	54	.syntax unified\r
	55	\r
	56	.type ASM_PFX(InternalMemScanMem8), %function\r
	57	ASM_GLOBAL ASM_PFX(InternalMemScanMem8)\r
	58	ASM_PFX(InternalMemScanMem8):\r
	59	// r0 = start of memory to scan\r
	60	// r1 = length\r
	61	// r2 = character to look for\r
	62	// returns r0 = pointer to character or NULL if not found\r
	63	uxtb r2, r2 // Don't think we can trust the caller to actually pass a char\r
	64	\r
65	cmp r1, #16 // If it's short don't bother with anything clever\r
66	blt 20f\r
67	\r
68	tst r0, #7 // If it's already aligned skip the next bit\r
69	beq 10f\r
70	\r
71	// Work up to an aligned point\r
72	5:\r
73	ldrb r3, [r0],#1\r
74	subs r1, r1, #1\r
75	cmp r3, r2\r
76	beq 50f // If it matches exit found\r
77	tst r0, #7\r
78	bne 5b // If not aligned yet then do next byte\r
79	\r
80	10:\r
81	// At this point, we are aligned, we know we have at least 8 bytes to work with\r
82	push {r4-r7}\r
83	orr r2, r2, r2, lsl #8 // expand the match word across to all bytes\r
84	orr r2, r2, r2, lsl #16\r
85	bic r4, r1, #7 // Number of double words to work with\r
86	mvns r7, #0 // all F's\r
87	movs r3, #0\r
88	\r
89	15:\r
90	ldmia r0!, {r5,r6}\r
91	subs r4, r4, #8\r
92	eor r5, r5, r2 // Get it so that r5,r6 have 00's where the bytes match the target\r
93	eor r6, r6, r2\r
94	uadd8 r5, r5, r7 // Parallel add 0xff - sets the GE bits for anything that wasn't 0\r
95	sel r5, r3, r7 // bytes are 00 for none-00 bytes, or ff for 00 bytes - NOTE INVERSION\r
96	uadd8 r6, r6, r7 // Parallel add 0xff - sets the GE bits for anything that wasn't 0\r
97	sel r6, r5, r7 // chained....bytes are 00 for none-00 bytes, or ff for 00 bytes - NOTE INVERSION\r
98	cbnz r6, 60f\r
99	bne 15b // (Flags from the subs above) If not run out of bytes then go around again\r
100	\r
101	pop {r4-r7}\r
102	and r2, r2, #0xff // Get r2 back to a single character from the expansion above\r
103	and r1, r1, #7 // Leave the count remaining as the number after the double words have been done\r
104	\r
105	20:\r
106	cbz r1, 40f // 0 length or hit the end already then not found\r
107	\r
108	21: // Post aligned section, or just a short call\r
109	ldrb r3, [r0], #1\r
110	subs r1, r1, #1\r
111	eor r3, r3, r2 // r3 = 0 if match - doesn't break flags from sub\r
112	cbz r3, 50f\r
113	bne 21b // on r1 flags\r
114	\r
115	40:\r
116	movs r0, #0 // not found\r
117	bx lr\r
118	\r
119	50:\r
120	subs r0, r0, #1 // found\r
121	bx lr\r
122	\r
123	60: // We're here because the fast path found a hit - now we have to track down exactly which word it was\r
124	// r0 points to the start of the double word after the one that was tested\r
125	// r5 has the 00/ff pattern for the first word, r6 has the chained value\r
126	cmp r5, #0\r
127	itte eq\r
128	moveq r5, r6 // the end is in the 2nd word\r
129	subeq r0, r0, #3 // Points to 2nd byte of 2nd word\r
130	subne r0, r0, #7 // or 2nd byte of 1st word\r
131	\r
132	// r0 currently points to the 3rd byte of the word containing the hit\r
133	tst r5, #CHARTSTMASK(0) // 1st character\r
134	bne 61f\r
135	adds r0, r0, #1\r
136	tst r5, #CHARTSTMASK(1) // 2nd character\r
c4f63707 AB	137	bne 61f\r
	138	adds r0, r0 ,#1\r
	139	tst r5, #(3 << 15) // 2nd & 3rd character\r
a37f6605	140	// If not the 3rd must be the last one\r
c4f63707 AB	141	it eq\r
c4f63707 AB	142	addeq.n r0, r0, #1\r
a37f6605 AB	143	\r
	144	61:\r
	145	pop {r4-r7}\r
	146	subs r0, r0, #1\r
	147	bx lr\r