include/asm/hash.h

1 /* SPDX-License-Identifier: GPL-2.0 */
6  * HP-PA only implements integer multiply in the FPU.  However, for
7  * integer multiplies by constant, it has a number of shift-and-add
8  * (but no shift-and-subtract, sigh!) instructions that a compiler
20  * PA7100 pairing rules.  This is an in-order 2-way superscalar processor.
22  * but other than that, simple ALU ops (including shift-and-add by up
25  * PA8xxx processors also dual-issue ALU instructions, although with
28  * This 6-step sequence was found by Yevgen Voronenko's implementation
43 	/* Phase 2: Return (b<<11) + (c<<6) + (a<<3) - c */  in __hash_32()
45 	a += c << 3;	b -= c;  in __hash_32()
54  * Finding a good shift-and-add chain for GOLDEN_RATIO_64 is tricky,
59  * However, Jason Thong pointed out a work-around.  The Hcub software
61  * constant multiplication, and is good at finding shift-and-add chains
68  * you can see the non-zero bits are divided into several well-separated
72  * Dividing the input into 2, 3 or 4 blocks, Hcub can find solutions
76  * Using just two large blocks, 0xC3910C8D << 31 in the high bits,
80  * The high bits are a larger number and more work to compute, as well
91  * This prevents it from mis-optimizing certain sequences.
93  * Given "x <<= 19; y += x; z += x << 1;", GCC will turn this into
94  * "y += x << 19; z += x << 20;" even though the latter sequence needs
98  * usefully portable across all GCC platforms, and so can be test-compiled
99  * on non-PA systems.
101  * In two places, additional unused input dependencies are added.  This
103  * Because the PA-8xxx is out of order, I'm not sure how much this matters,
110  * optimized shift-and-add sequence.
127 		asm("" : "=q" (bits) : "0" (64 - bits));  in hash_64()
129 		bits = 64 - bits;  in hash_64()
135 	d = a - d;		_ASSIGN(a, a << 4, "X" (d));  in hash_64()
137 	d -= c;			c += a << 1;  in hash_64()
143 #undef _ASSIGN	/* We're a widely-used header file, so don't litter! */