Lines Matching +full:full +full:- +full:bit

30  * (from the AES-NI instructions).
42 	 * Bit mask for features in ECX:  in pclmul_supported()
58  * GHASH is defined over elements of GF(2^128) with "full little-endian"
61  * x86 is "mixed little-endian": bytes are ordered from least to most
62  * significant, but bits within a byte are in most-to-least significant
63  * order. Going to full little-endian representation would require
66  * Instead, we go to full big-endian representation, by swapping bytes
69  * can use a full big-endian representation because in a carryless
70  * multiplication, we have a nice bit reversal property:
74  * So by using full big-endian, we still get the right result, except
75  * that it is right-shifted by 1 bit. The left-shift is relatively
79  * Since SSE2 opcodes do not have facilities for shitfting full 128-bit
80  * values with bit precision, we have to break down values into 64-bit
85  * Byte-swap a complete 128-bit value. This normally uses
117  * potentially a bit slower because the compiler does not understand
145  * From a 128-bit value kw, compute kx as the XOR of the two 64-bit
153  * Combine two 64-bit values (k0:k1) into a 128-bit (kw) value and
162  * Left-shift by 1 bit a 256-bit value (in four 64-bit words).
178  * Perform reduction in GF(2^128). The 256-bit value is in x0..x3;
243 	 * and contains num4 blocks of 64-byte values. Second chunk  in br_ghash_pclmul()
244 	 * starts at buf2 and contains num1 blocks of 16-byte values.  in br_ghash_pclmul()
254 		memset(tmp + len, 0, (num1 << 4) - len);  in br_ghash_pclmul()
300 		while (num4 -- > 0) {  in br_ghash_pclmul()
351 	while (num1 -- > 0) {  in br_ghash_pclmul()