Text file src/vendor/golang.org/x/crypto/internal/poly1305/sum_s390x.s

     1  // Copyright 2018 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  //go:build gc && !purego
     6  
     7  #include "textflag.h"
     8  
     9  // This implementation of Poly1305 uses the vector facility (vx)
    10  // to process up to 2 blocks (32 bytes) per iteration using an
    11  // algorithm based on the one described in:
    12  //
    13  // NEON crypto, Daniel J. Bernstein & Peter Schwabe
    14  // https://cryptojedi.org/papers/neoncrypto-20120320.pdf
    15  //
    16  // This algorithm uses 5 26-bit limbs to represent a 130-bit
    17  // value. These limbs are, for the most part, zero extended and
    18  // placed into 64-bit vector register elements. Each vector
    19  // register is 128-bits wide and so holds 2 of these elements.
    20  // Using 26-bit limbs allows us plenty of headroom to accommodate
    21  // accumulations before and after multiplication without
    22  // overflowing either 32-bits (before multiplication) or 64-bits
    23  // (after multiplication).
    24  //
    25  // In order to parallelise the operations required to calculate
    26  // the sum we use two separate accumulators and then sum those
    27  // in an extra final step. For compatibility with the generic
    28  // implementation we perform this summation at the end of every
    29  // updateVX call.
    30  //
    31  // To use two accumulators we must multiply the message blocks
    32  // by r² rather than r. Only the final message block should be
    33  // multiplied by r.
    34  //
    35  // Example:
    36  //
    37  // We want to calculate the sum (h) for a 64 byte message (m):
    38  //
    39  //   h = m[0:16]r⁴ + m[16:32]r³ + m[32:48]r² + m[48:64]r
    40  //
    41  // To do this we split the calculation into the even indices
    42  // and odd indices of the message. These form our SIMD 'lanes':
    43  //
    44  //   h = m[ 0:16]r⁴ + m[32:48]r² +   <- lane 0
    45  //       m[16:32]r³ + m[48:64]r      <- lane 1
    46  //
    47  // To calculate this iteratively we refactor so that both lanes
    48  // are written in terms of r² and r:
    49  //
    50  //   h = (m[ 0:16]r² + m[32:48])r² + <- lane 0
    51  //       (m[16:32]r² + m[48:64])r    <- lane 1
    52  //                ^             ^
    53  //                |             coefficients for second iteration
    54  //                coefficients for first iteration
    55  //
    56  // So in this case we would have two iterations. In the first
    57  // both lanes are multiplied by r². In the second only the
    58  // first lane is multiplied by r² and the second lane is
    59  // instead multiplied by r. This gives use the odd and even
    60  // powers of r that we need from the original equation.
    61  //
    62  // Notation:
    63  //
    64  //   h - accumulator
    65  //   r - key
    66  //   m - message
    67  //
    68  //   [a, b]       - SIMD register holding two 64-bit values
    69  //   [a, b, c, d] - SIMD register holding four 32-bit values
    70  //   xᵢ[n]        - limb n of variable x with bit width i
    71  //
    72  // Limbs are expressed in little endian order, so for 26-bit
    73  // limbs x₂₆[4] will be the most significant limb and x₂₆[0]
    74  // will be the least significant limb.
    75  
    76  // masking constants
    77  #define MOD24 V0 // [0x0000000000ffffff, 0x0000000000ffffff] - mask low 24-bits
    78  #define MOD26 V1 // [0x0000000003ffffff, 0x0000000003ffffff] - mask low 26-bits
    79  
    80  // expansion constants (see EXPAND macro)
    81  #define EX0 V2
    82  #define EX1 V3
    83  #define EX2 V4
    84  
    85  // key (r², r or 1 depending on context)
    86  #define R_0 V5
    87  #define R_1 V6
    88  #define R_2 V7
    89  #define R_3 V8
    90  #define R_4 V9
    91  
    92  // precalculated coefficients (5r², 5r or 0 depending on context)
    93  #define R5_1 V10
    94  #define R5_2 V11
    95  #define R5_3 V12
    96  #define R5_4 V13
    97  
    98  // message block (m)
    99  #define M_0 V14
   100  #define M_1 V15
   101  #define M_2 V16
   102  #define M_3 V17
   103  #define M_4 V18
   104  
   105  // accumulator (h)
   106  #define H_0 V19
   107  #define H_1 V20
   108  #define H_2 V21
   109  #define H_3 V22
   110  #define H_4 V23
   111  
   112  // temporary registers (for short-lived values)
   113  #define T_0 V24
   114  #define T_1 V25
   115  #define T_2 V26
   116  #define T_3 V27
   117  #define T_4 V28
   118  
   119  GLOBL ·constants<>(SB), RODATA, $0x30
   120  // EX0
   121  DATA ·constants<>+0x00(SB)/8, $0x0006050403020100
   122  DATA ·constants<>+0x08(SB)/8, $0x1016151413121110
   123  // EX1
   124  DATA ·constants<>+0x10(SB)/8, $0x060c0b0a09080706
   125  DATA ·constants<>+0x18(SB)/8, $0x161c1b1a19181716
   126  // EX2
   127  DATA ·constants<>+0x20(SB)/8, $0x0d0d0d0d0d0f0e0d
   128  DATA ·constants<>+0x28(SB)/8, $0x1d1d1d1d1d1f1e1d
   129  
   130  // MULTIPLY multiplies each lane of f and g, partially reduced
   131  // modulo 2¹³⁰ - 5. The result, h, consists of partial products
   132  // in each lane that need to be reduced further to produce the
   133  // final result.
   134  //
   135  //   h₁₃₀ = (f₁₃₀g₁₃₀) % 2¹³⁰ + (5f₁₃₀g₁₃₀) / 2¹³⁰
   136  //
   137  // Note that the multiplication by 5 of the high bits is
   138  // achieved by precalculating the multiplication of four of the
   139  // g coefficients by 5. These are g51-g54.
   140  #define MULTIPLY(f0, f1, f2, f3, f4, g0, g1, g2, g3, g4, g51, g52, g53, g54, h0, h1, h2, h3, h4) \
   141  	VMLOF  f0, g0, h0        \
   142  	VMLOF  f0, g3, h3        \
   143  	VMLOF  f0, g1, h1        \
   144  	VMLOF  f0, g4, h4        \
   145  	VMLOF  f0, g2, h2        \
   146  	VMLOF  f1, g54, T_0      \
   147  	VMLOF  f1, g2, T_3       \
   148  	VMLOF  f1, g0, T_1       \
   149  	VMLOF  f1, g3, T_4       \
   150  	VMLOF  f1, g1, T_2       \
   151  	VMALOF f2, g53, h0, h0   \
   152  	VMALOF f2, g1, h3, h3    \
   153  	VMALOF f2, g54, h1, h1   \
   154  	VMALOF f2, g2, h4, h4    \
   155  	VMALOF f2, g0, h2, h2    \
   156  	VMALOF f3, g52, T_0, T_0 \
   157  	VMALOF f3, g0, T_3, T_3  \
   158  	VMALOF f3, g53, T_1, T_1 \
   159  	VMALOF f3, g1, T_4, T_4  \
   160  	VMALOF f3, g54, T_2, T_2 \
   161  	VMALOF f4, g51, h0, h0   \
   162  	VMALOF f4, g54, h3, h3   \
   163  	VMALOF f4, g52, h1, h1   \
   164  	VMALOF f4, g0, h4, h4    \
   165  	VMALOF f4, g53, h2, h2   \
   166  	VAG    T_0, h0, h0       \
   167  	VAG    T_3, h3, h3       \
   168  	VAG    T_1, h1, h1       \
   169  	VAG    T_4, h4, h4       \
   170  	VAG    T_2, h2, h2
   171  
   172  // REDUCE performs the following carry operations in four
   173  // stages, as specified in Bernstein & Schwabe:
   174  //
   175  //   1: h₂₆[0]->h₂₆[1] h₂₆[3]->h₂₆[4]
   176  //   2: h₂₆[1]->h₂₆[2] h₂₆[4]->h₂₆[0]
   177  //   3: h₂₆[0]->h₂₆[1] h₂₆[2]->h₂₆[3]
   178  //   4: h₂₆[3]->h₂₆[4]
   179  //
   180  // The result is that all of the limbs are limited to 26-bits
   181  // except for h₂₆[1] and h₂₆[4] which are limited to 27-bits.
   182  //
   183  // Note that although each limb is aligned at 26-bit intervals
   184  // they may contain values that exceed 2²⁶ - 1, hence the need
   185  // to carry the excess bits in each limb.
   186  #define REDUCE(h0, h1, h2, h3, h4) \
   187  	VESRLG $26, h0, T_0  \
   188  	VESRLG $26, h3, T_1  \
   189  	VN     MOD26, h0, h0 \
   190  	VN     MOD26, h3, h3 \
   191  	VAG    T_0, h1, h1   \
   192  	VAG    T_1, h4, h4   \
   193  	VESRLG $26, h1, T_2  \
   194  	VESRLG $26, h4, T_3  \
   195  	VN     MOD26, h1, h1 \
   196  	VN     MOD26, h4, h4 \
   197  	VESLG  $2, T_3, T_4  \
   198  	VAG    T_3, T_4, T_4 \
   199  	VAG    T_2, h2, h2   \
   200  	VAG    T_4, h0, h0   \
   201  	VESRLG $26, h2, T_0  \
   202  	VESRLG $26, h0, T_1  \
   203  	VN     MOD26, h2, h2 \
   204  	VN     MOD26, h0, h0 \
   205  	VAG    T_0, h3, h3   \
   206  	VAG    T_1, h1, h1   \
   207  	VESRLG $26, h3, T_2  \
   208  	VN     MOD26, h3, h3 \
   209  	VAG    T_2, h4, h4
   210  
   211  // EXPAND splits the 128-bit little-endian values in0 and in1
   212  // into 26-bit big-endian limbs and places the results into
   213  // the first and second lane of d₂₆[0:4] respectively.
   214  //
   215  // The EX0, EX1 and EX2 constants are arrays of byte indices
   216  // for permutation. The permutation both reverses the bytes
   217  // in the input and ensures the bytes are copied into the
   218  // destination limb ready to be shifted into their final
   219  // position.
   220  #define EXPAND(in0, in1, d0, d1, d2, d3, d4) \
   221  	VPERM  in0, in1, EX0, d0 \
   222  	VPERM  in0, in1, EX1, d2 \
   223  	VPERM  in0, in1, EX2, d4 \
   224  	VESRLG $26, d0, d1       \
   225  	VESRLG $30, d2, d3       \
   226  	VESRLG $4, d2, d2        \
   227  	VN     MOD26, d0, d0     \ // [in0₂₆[0], in1₂₆[0]]
   228  	VN     MOD26, d3, d3     \ // [in0₂₆[3], in1₂₆[3]]
   229  	VN     MOD26, d1, d1     \ // [in0₂₆[1], in1₂₆[1]]
   230  	VN     MOD24, d4, d4     \ // [in0₂₆[4], in1₂₆[4]]
   231  	VN     MOD26, d2, d2     // [in0₂₆[2], in1₂₆[2]]
   232  
   233  // func updateVX(state *macState, msg []byte)
   234  TEXT ·updateVX(SB), NOSPLIT, $0
   235  	MOVD state+0(FP), R1
   236  	LMG  msg+8(FP), R2, R3 // R2=msg_base, R3=msg_len
   237  
   238  	// load EX0, EX1 and EX2
   239  	MOVD $·constants<>(SB), R5
   240  	VLM  (R5), EX0, EX2
   241  
   242  	// generate masks
   243  	VGMG $(64-24), $63, MOD24 // [0x00ffffff, 0x00ffffff]
   244  	VGMG $(64-26), $63, MOD26 // [0x03ffffff, 0x03ffffff]
   245  
   246  	// load h (accumulator) and r (key) from state
   247  	VZERO T_1               // [0, 0]
   248  	VL    0(R1), T_0        // [h₆₄[0], h₆₄[1]]
   249  	VLEG  $0, 16(R1), T_1   // [h₆₄[2], 0]
   250  	VL    24(R1), T_2       // [r₆₄[0], r₆₄[1]]
   251  	VPDI  $0, T_0, T_2, T_3 // [h₆₄[0], r₆₄[0]]
   252  	VPDI  $5, T_0, T_2, T_4 // [h₆₄[1], r₆₄[1]]
   253  
   254  	// unpack h and r into 26-bit limbs
   255  	// note: h₆₄[2] may have the low 3 bits set, so h₂₆[4] is a 27-bit value
   256  	VN     MOD26, T_3, H_0            // [h₂₆[0], r₂₆[0]]
   257  	VZERO  H_1                        // [0, 0]
   258  	VZERO  H_3                        // [0, 0]
   259  	VGMG   $(64-12-14), $(63-12), T_0 // [0x03fff000, 0x03fff000] - 26-bit mask with low 12 bits masked out
   260  	VESLG  $24, T_1, T_1              // [h₆₄[2]<<24, 0]
   261  	VERIMG $-26&63, T_3, MOD26, H_1   // [h₂₆[1], r₂₆[1]]
   262  	VESRLG $+52&63, T_3, H_2          // [h₂₆[2], r₂₆[2]] - low 12 bits only
   263  	VERIMG $-14&63, T_4, MOD26, H_3   // [h₂₆[1], r₂₆[1]]
   264  	VESRLG $40, T_4, H_4              // [h₂₆[4], r₂₆[4]] - low 24 bits only
   265  	VERIMG $+12&63, T_4, T_0, H_2     // [h₂₆[2], r₂₆[2]] - complete
   266  	VO     T_1, H_4, H_4              // [h₂₆[4], r₂₆[4]] - complete
   267  
   268  	// replicate r across all 4 vector elements
   269  	VREPF $3, H_0, R_0 // [r₂₆[0], r₂₆[0], r₂₆[0], r₂₆[0]]
   270  	VREPF $3, H_1, R_1 // [r₂₆[1], r₂₆[1], r₂₆[1], r₂₆[1]]
   271  	VREPF $3, H_2, R_2 // [r₂₆[2], r₂₆[2], r₂₆[2], r₂₆[2]]
   272  	VREPF $3, H_3, R_3 // [r₂₆[3], r₂₆[3], r₂₆[3], r₂₆[3]]
   273  	VREPF $3, H_4, R_4 // [r₂₆[4], r₂₆[4], r₂₆[4], r₂₆[4]]
   274  
   275  	// zero out lane 1 of h
   276  	VLEIG $1, $0, H_0 // [h₂₆[0], 0]
   277  	VLEIG $1, $0, H_1 // [h₂₆[1], 0]
   278  	VLEIG $1, $0, H_2 // [h₂₆[2], 0]
   279  	VLEIG $1, $0, H_3 // [h₂₆[3], 0]
   280  	VLEIG $1, $0, H_4 // [h₂₆[4], 0]
   281  
   282  	// calculate 5r (ignore least significant limb)
   283  	VREPIF $5, T_0
   284  	VMLF   T_0, R_1, R5_1 // [5r₂₆[1], 5r₂₆[1], 5r₂₆[1], 5r₂₆[1]]
   285  	VMLF   T_0, R_2, R5_2 // [5r₂₆[2], 5r₂₆[2], 5r₂₆[2], 5r₂₆[2]]
   286  	VMLF   T_0, R_3, R5_3 // [5r₂₆[3], 5r₂₆[3], 5r₂₆[3], 5r₂₆[3]]
   287  	VMLF   T_0, R_4, R5_4 // [5r₂₆[4], 5r₂₆[4], 5r₂₆[4], 5r₂₆[4]]
   288  
   289  	// skip r² calculation if we are only calculating one block
   290  	CMPBLE R3, $16, skip
   291  
   292  	// calculate r²
   293  	MULTIPLY(R_0, R_1, R_2, R_3, R_4, R_0, R_1, R_2, R_3, R_4, R5_1, R5_2, R5_3, R5_4, M_0, M_1, M_2, M_3, M_4)
   294  	REDUCE(M_0, M_1, M_2, M_3, M_4)
   295  	VGBM   $0x0f0f, T_0
   296  	VERIMG $0, M_0, T_0, R_0 // [r₂₆[0], r²₂₆[0], r₂₆[0], r²₂₆[0]]
   297  	VERIMG $0, M_1, T_0, R_1 // [r₂₆[1], r²₂₆[1], r₂₆[1], r²₂₆[1]]
   298  	VERIMG $0, M_2, T_0, R_2 // [r₂₆[2], r²₂₆[2], r₂₆[2], r²₂₆[2]]
   299  	VERIMG $0, M_3, T_0, R_3 // [r₂₆[3], r²₂₆[3], r₂₆[3], r²₂₆[3]]
   300  	VERIMG $0, M_4, T_0, R_4 // [r₂₆[4], r²₂₆[4], r₂₆[4], r²₂₆[4]]
   301  
   302  	// calculate 5r² (ignore least significant limb)
   303  	VREPIF $5, T_0
   304  	VMLF   T_0, R_1, R5_1 // [5r₂₆[1], 5r²₂₆[1], 5r₂₆[1], 5r²₂₆[1]]
   305  	VMLF   T_0, R_2, R5_2 // [5r₂₆[2], 5r²₂₆[2], 5r₂₆[2], 5r²₂₆[2]]
   306  	VMLF   T_0, R_3, R5_3 // [5r₂₆[3], 5r²₂₆[3], 5r₂₆[3], 5r²₂₆[3]]
   307  	VMLF   T_0, R_4, R5_4 // [5r₂₆[4], 5r²₂₆[4], 5r₂₆[4], 5r²₂₆[4]]
   308  
   309  loop:
   310  	CMPBLE R3, $32, b2 // 2 or fewer blocks remaining, need to change key coefficients
   311  
   312  	// load next 2 blocks from message
   313  	VLM (R2), T_0, T_1
   314  
   315  	// update message slice
   316  	SUB  $32, R3
   317  	MOVD $32(R2), R2
   318  
   319  	// unpack message blocks into 26-bit big-endian limbs
   320  	EXPAND(T_0, T_1, M_0, M_1, M_2, M_3, M_4)
   321  
   322  	// add 2¹²⁸ to each message block value
   323  	VLEIB $4, $1, M_4
   324  	VLEIB $12, $1, M_4
   325  
   326  multiply:
   327  	// accumulate the incoming message
   328  	VAG H_0, M_0, M_0
   329  	VAG H_3, M_3, M_3
   330  	VAG H_1, M_1, M_1
   331  	VAG H_4, M_4, M_4
   332  	VAG H_2, M_2, M_2
   333  
   334  	// multiply the accumulator by the key coefficient
   335  	MULTIPLY(M_0, M_1, M_2, M_3, M_4, R_0, R_1, R_2, R_3, R_4, R5_1, R5_2, R5_3, R5_4, H_0, H_1, H_2, H_3, H_4)
   336  
   337  	// carry and partially reduce the partial products
   338  	REDUCE(H_0, H_1, H_2, H_3, H_4)
   339  
   340  	CMPBNE R3, $0, loop
   341  
   342  finish:
   343  	// sum lane 0 and lane 1 and put the result in lane 1
   344  	VZERO  T_0
   345  	VSUMQG H_0, T_0, H_0
   346  	VSUMQG H_3, T_0, H_3
   347  	VSUMQG H_1, T_0, H_1
   348  	VSUMQG H_4, T_0, H_4
   349  	VSUMQG H_2, T_0, H_2
   350  
   351  	// reduce again after summation
   352  	// TODO(mundaym): there might be a more efficient way to do this
   353  	// now that we only have 1 active lane. For example, we could
   354  	// simultaneously pack the values as we reduce them.
   355  	REDUCE(H_0, H_1, H_2, H_3, H_4)
   356  
   357  	// carry h[1] through to h[4] so that only h[4] can exceed 2²⁶ - 1
   358  	// TODO(mundaym): in testing this final carry was unnecessary.
   359  	// Needs a proof before it can be removed though.
   360  	VESRLG $26, H_1, T_1
   361  	VN     MOD26, H_1, H_1
   362  	VAQ    T_1, H_2, H_2
   363  	VESRLG $26, H_2, T_2
   364  	VN     MOD26, H_2, H_2
   365  	VAQ    T_2, H_3, H_3
   366  	VESRLG $26, H_3, T_3
   367  	VN     MOD26, H_3, H_3
   368  	VAQ    T_3, H_4, H_4
   369  
   370  	// h is now < 2(2¹³⁰ - 5)
   371  	// Pack each lane in h₂₆[0:4] into h₁₂₈[0:1].
   372  	VESLG $26, H_1, H_1
   373  	VESLG $26, H_3, H_3
   374  	VO    H_0, H_1, H_0
   375  	VO    H_2, H_3, H_2
   376  	VESLG $4, H_2, H_2
   377  	VLEIB $7, $48, H_1
   378  	VSLB  H_1, H_2, H_2
   379  	VO    H_0, H_2, H_0
   380  	VLEIB $7, $104, H_1
   381  	VSLB  H_1, H_4, H_3
   382  	VO    H_3, H_0, H_0
   383  	VLEIB $7, $24, H_1
   384  	VSRLB H_1, H_4, H_1
   385  
   386  	// update state
   387  	VSTEG $1, H_0, 0(R1)
   388  	VSTEG $0, H_0, 8(R1)
   389  	VSTEG $1, H_1, 16(R1)
   390  	RET
   391  
   392  b2:  // 2 or fewer blocks remaining
   393  	CMPBLE R3, $16, b1
   394  
   395  	// Load the 2 remaining blocks (17-32 bytes remaining).
   396  	MOVD $-17(R3), R0    // index of final byte to load modulo 16
   397  	VL   (R2), T_0       // load full 16 byte block
   398  	VLL  R0, 16(R2), T_1 // load final (possibly partial) block and pad with zeros to 16 bytes
   399  
   400  	// The Poly1305 algorithm requires that a 1 bit be appended to
   401  	// each message block. If the final block is less than 16 bytes
   402  	// long then it is easiest to insert the 1 before the message
   403  	// block is split into 26-bit limbs. If, on the other hand, the
   404  	// final message block is 16 bytes long then we append the 1 bit
   405  	// after expansion as normal.
   406  	MOVBZ  $1, R0
   407  	MOVD   $-16(R3), R3   // index of byte in last block to insert 1 at (could be 16)
   408  	CMPBEQ R3, $16, 2(PC) // skip the insertion if the final block is 16 bytes long
   409  	VLVGB  R3, R0, T_1    // insert 1 into the byte at index R3
   410  
   411  	// Split both blocks into 26-bit limbs in the appropriate lanes.
   412  	EXPAND(T_0, T_1, M_0, M_1, M_2, M_3, M_4)
   413  
   414  	// Append a 1 byte to the end of the second to last block.
   415  	VLEIB $4, $1, M_4
   416  
   417  	// Append a 1 byte to the end of the last block only if it is a
   418  	// full 16 byte block.
   419  	CMPBNE R3, $16, 2(PC)
   420  	VLEIB  $12, $1, M_4
   421  
   422  	// Finally, set up the coefficients for the final multiplication.
   423  	// We have previously saved r and 5r in the 32-bit even indexes
   424  	// of the R_[0-4] and R5_[1-4] coefficient registers.
   425  	//
   426  	// We want lane 0 to be multiplied by r² so that can be kept the
   427  	// same. We want lane 1 to be multiplied by r so we need to move
   428  	// the saved r value into the 32-bit odd index in lane 1 by
   429  	// rotating the 64-bit lane by 32.
   430  	VGBM   $0x00ff, T_0         // [0, 0xffffffffffffffff] - mask lane 1 only
   431  	VERIMG $32, R_0, T_0, R_0   // [_,  r²₂₆[0], _,  r₂₆[0]]
   432  	VERIMG $32, R_1, T_0, R_1   // [_,  r²₂₆[1], _,  r₂₆[1]]
   433  	VERIMG $32, R_2, T_0, R_2   // [_,  r²₂₆[2], _,  r₂₆[2]]
   434  	VERIMG $32, R_3, T_0, R_3   // [_,  r²₂₆[3], _,  r₂₆[3]]
   435  	VERIMG $32, R_4, T_0, R_4   // [_,  r²₂₆[4], _,  r₂₆[4]]
   436  	VERIMG $32, R5_1, T_0, R5_1 // [_, 5r²₂₆[1], _, 5r₂₆[1]]
   437  	VERIMG $32, R5_2, T_0, R5_2 // [_, 5r²₂₆[2], _, 5r₂₆[2]]
   438  	VERIMG $32, R5_3, T_0, R5_3 // [_, 5r²₂₆[3], _, 5r₂₆[3]]
   439  	VERIMG $32, R5_4, T_0, R5_4 // [_, 5r²₂₆[4], _, 5r₂₆[4]]
   440  
   441  	MOVD $0, R3
   442  	BR   multiply
   443  
   444  skip:
   445  	CMPBEQ R3, $0, finish
   446  
   447  b1:  // 1 block remaining
   448  
   449  	// Load the final block (1-16 bytes). This will be placed into
   450  	// lane 0.
   451  	MOVD $-1(R3), R0
   452  	VLL  R0, (R2), T_0 // pad to 16 bytes with zeros
   453  
   454  	// The Poly1305 algorithm requires that a 1 bit be appended to
   455  	// each message block. If the final block is less than 16 bytes
   456  	// long then it is easiest to insert the 1 before the message
   457  	// block is split into 26-bit limbs. If, on the other hand, the
   458  	// final message block is 16 bytes long then we append the 1 bit
   459  	// after expansion as normal.
   460  	MOVBZ  $1, R0
   461  	CMPBEQ R3, $16, 2(PC)
   462  	VLVGB  R3, R0, T_0
   463  
   464  	// Set the message block in lane 1 to the value 0 so that it
   465  	// can be accumulated without affecting the final result.
   466  	VZERO T_1
   467  
   468  	// Split the final message block into 26-bit limbs in lane 0.
   469  	// Lane 1 will be contain 0.
   470  	EXPAND(T_0, T_1, M_0, M_1, M_2, M_3, M_4)
   471  
   472  	// Append a 1 byte to the end of the last block only if it is a
   473  	// full 16 byte block.
   474  	CMPBNE R3, $16, 2(PC)
   475  	VLEIB  $4, $1, M_4
   476  
   477  	// We have previously saved r and 5r in the 32-bit even indexes
   478  	// of the R_[0-4] and R5_[1-4] coefficient registers.
   479  	//
   480  	// We want lane 0 to be multiplied by r so we need to move the
   481  	// saved r value into the 32-bit odd index in lane 0. We want
   482  	// lane 1 to be set to the value 1. This makes multiplication
   483  	// a no-op. We do this by setting lane 1 in every register to 0
   484  	// and then just setting the 32-bit index 3 in R_0 to 1.
   485  	VZERO T_0
   486  	MOVD  $0, R0
   487  	MOVD  $0x10111213, R12
   488  	VLVGP R12, R0, T_1         // [_, 0x10111213, _, 0x00000000]
   489  	VPERM T_0, R_0, T_1, R_0   // [_,  r₂₆[0], _, 0]
   490  	VPERM T_0, R_1, T_1, R_1   // [_,  r₂₆[1], _, 0]
   491  	VPERM T_0, R_2, T_1, R_2   // [_,  r₂₆[2], _, 0]
   492  	VPERM T_0, R_3, T_1, R_3   // [_,  r₂₆[3], _, 0]
   493  	VPERM T_0, R_4, T_1, R_4   // [_,  r₂₆[4], _, 0]
   494  	VPERM T_0, R5_1, T_1, R5_1 // [_, 5r₂₆[1], _, 0]
   495  	VPERM T_0, R5_2, T_1, R5_2 // [_, 5r₂₆[2], _, 0]
   496  	VPERM T_0, R5_3, T_1, R5_3 // [_, 5r₂₆[3], _, 0]
   497  	VPERM T_0, R5_4, T_1, R5_4 // [_, 5r₂₆[4], _, 0]
   498  
   499  	// Set the value of lane 1 to be 1.
   500  	VLEIF $3, $1, R_0 // [_,  r₂₆[0], _, 1]
   501  
   502  	MOVD $0, R3
   503  	BR   multiply
   504  

View as plain text