github.com/sandwichdev/go-internals@v0.0.0-20210605002614-12311ac6b2c5/bytealg/count_ppc64x.s (about)

     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  // +build ppc64le ppc64
     6  
     7  #include "go_asm.h"
     8  #include "textflag.h"
     9  
    10  TEXT ·Count(SB), NOSPLIT|NOFRAME, $0-40
    11  	MOVD  b_base+0(FP), R3    // R3 = byte array pointer
    12  	MOVD  b_len+8(FP), R4     // R4 = length
    13  	MOVBZ c+24(FP), R5        // R5 = byte
    14  	MOVD  $ret+32(FP), R14    // R14 = &ret
    15  	BR    countbytebody<>(SB)
    16  
    17  TEXT ·CountString(SB), NOSPLIT|NOFRAME, $0-32
    18  	MOVD  s_base+0(FP), R3    // R3 = string
    19  	MOVD  s_len+8(FP), R4     // R4 = length
    20  	MOVBZ c+16(FP), R5        // R5 = byte
    21  	MOVD  $ret+24(FP), R14    // R14 = &ret
    22  	BR    countbytebody<>(SB)
    23  
    24  // R3: addr of string
    25  // R4: len of string
    26  // R5: byte to count
    27  // R14: addr for return value
    28  // endianness shouldn't matter since we are just counting and order
    29  // is irrelevant
    30  TEXT countbytebody<>(SB), NOSPLIT|NOFRAME, $0-0
    31  	DCBT (R3)    // Prepare cache line.
    32  	MOVD R0, R18 // byte count
    33  	MOVD R3, R19 // Save base address for calculating the index later.
    34  	MOVD R4, R16
    35  
    36  	MOVD   R5, R6
    37  	RLDIMI $8, R6, $48, R6
    38  	RLDIMI $16, R6, $32, R6
    39  	RLDIMI $32, R6, $0, R6  // fill reg with the byte to count
    40  
    41  	VSPLTISW $3, V4     // used for shift
    42  	MTVRD    R6, V1     // move compare byte
    43  	VSPLTB   $7, V1, V1 // replicate byte across V1
    44  
    45  	CMPU   R4, $32          // Check if it's a small string (<32 bytes)
    46  	BLT    tail             // Jump to the small string case
    47  	XXLXOR VS37, VS37, VS37 // clear V5 (aka VS37) to use as accumulator
    48  
    49  cmploop:
    50  	LXVW4X (R3), VS32 // load bytes from string
    51  
    52  	// when the bytes match, the corresponding byte contains all 1s
    53  	VCMPEQUB V1, V0, V2     // compare bytes
    54  	VPOPCNTD V2, V3         // each double word contains its count
    55  	VADDUDM  V3, V5, V5     // accumulate bit count in each double word
    56  	ADD      $16, R3, R3    // increment pointer
    57  	SUB      $16, R16, R16  // remaining bytes
    58  	CMP      R16, $16       // at least 16 remaining?
    59  	BGE      cmploop
    60  	VSRD     V5, V4, V5     // shift by 3 to convert bits to bytes
    61  	VSLDOI   $8, V5, V5, V6 // get the double word values from vector
    62  	MFVSRD   V5, R9
    63  	MFVSRD   V6, R10
    64  	ADD      R9, R10, R9
    65  	ADD      R9, R18, R18
    66  
    67  tail:
    68  	CMP R16, $8 // 8 bytes left?
    69  	BLT small
    70  
    71  	MOVD    (R3), R12     // load 8 bytes
    72  	CMPB    R12, R6, R17  // compare bytes
    73  	POPCNTD R17, R15      // bit count
    74  	SRD     $3, R15, R15  // byte count
    75  	ADD     R15, R18, R18 // add to byte count
    76  
    77  next1:
    78  	ADD $8, R3, R3
    79  	SUB $8, R16, R16 // remaining bytes
    80  	BR  tail
    81  
    82  small:
    83  	CMP   $0, R16   // any remaining
    84  	BEQ   done
    85  	MOVBZ (R3), R12 // check each remaining byte
    86  	CMP   R12, R5
    87  	BNE   next2
    88  	ADD   $1, R18
    89  
    90  next2:
    91  	SUB $1, R16
    92  	ADD $1, R3  // inc address
    93  	BR  small
    94  
    95  done:
    96  	MOVD R18, (R14) // return count
    97  	RET