crypto: simplify amd64 asm for sha{1,256,512} a bit

Use constants directly, instead of loading address to e. g. AX
and using (AX). Shouldn't affect performance, but makes code a bit
nicer.

Change-Id: Ifa138e54d3d2b2f4ad71e4ef4b9368ea79eb30f4
Reviewed-on: https://go-review.googlesource.com/62010
Reviewed-by: Adam Langley <agl@golang.org>

diff --git a/src/crypto/sha1/sha1block_amd64.s b/src/crypto/sha1/sha1block_amd64.s
index 3adb6d2c32b8ef21edb7870f3f45a4f3e77217e1..a0032c4544095ef0e1d10f9d84a3f9eb93d8be87 100644 (file)
--- a/src/crypto/sha1/sha1block_amd64.s
+++ b/src/crypto/sha1/sha1block_amd64.s
@@ -1451,9 +1451,7 @@ TEXT ·blockAVX2(SB),$1408-32
        CMPQ    R13, R11
        CMOVQCC R8, R13
 
-       MOVQ    $BSWAP_SHUFB_CTL<>(SB), R8
-       VMOVDQU (R8), Y10
-       MOVQ    $K_XMM_AR<>(SB), R8 //restore R8
+       VMOVDQU BSWAP_SHUFB_CTL<>(SB), Y10
 
        CALC // RET is inside macros
 

diff --git a/src/crypto/sha256/sha256block_amd64.s b/src/crypto/sha256/sha256block_amd64.s
index d7ac1e53b3577eb5ebb5832fa2b7af71ba89f6df..f30f4829a6c1b0bb459fb50f184a4ddc9da4319a 100644 (file)
--- a/src/crypto/sha256/sha256block_amd64.s
+++ b/src/crypto/sha256/sha256block_amd64.s
@@ -213,13 +213,11 @@
 #define XFER_SIZE 2*64*4
 #define INP_END_SIZE 8
 #define INP_SIZE 8
-#define TMP_SIZE 4
 
 #define _XFER 0
 #define _INP_END _XFER + XFER_SIZE
 #define _INP _INP_END + INP_END_SIZE
-#define _TMP _INP + INP_SIZE
-#define STACK_SIZE _TMP + TMP_SIZE
+#define STACK_SIZE _INP + INP_SIZE
 
 #define ROUND_AND_SCHED_N_0(disp, a, b, c, d, e, f, g, h, XDWORD0, XDWORD1, XDWORD2, XDWORD3) \
        ;                                     \ // #############################  RND N + 0 ############################//
@@ -341,10 +339,7 @@
        VPXOR   XTMP2, XTMP4, XTMP4;         \ // XTMP4 = s1 {xBxA}
        XORL    g, y2;                       \ // y2 = CH = ((f^g)&e)^g                 // CH
        ;                                    \
-       MOVL    f, _TMP(SP);                 \
-       MOVQ    $shuff_00BA<>(SB), f;        \ // f is used to keep SHUF_00BA
-       VPSHUFB (f), XTMP4, XTMP4;           \ // XTMP4 = s1 {00BA}
-       MOVL    _TMP(SP), f;                 \ // f is restored
+       VPSHUFB shuff_00BA<>(SB), XTMP4, XTMP4;\ // XTMP4 = s1 {00BA}
        ;                                    \
        XORL    T1, y1;                      \ // y1 = (a>>22) ^ (a>>13)                // S0
        RORXL   $2, a, T1;                   \ // T1 = (a >> 2)                         // S0
@@ -398,10 +393,7 @@
        ;                                    \
        RORXL   $2, a, T1;                   \ // T1 = (a >> 2)                         // S0
        ;                                    \
-       MOVL    f, _TMP(SP);                 \ // Save f
-       MOVQ    $shuff_DC00<>(SB), f;        \ // SHUF_00DC
-       VPSHUFB (f), XTMP5, XTMP5;           \ // XTMP5 = s1 {DC00}
-       MOVL    _TMP(SP), f;                 \ // Restore f
+       VPSHUFB shuff_DC00<>(SB), XTMP5, XTMP5;\ // XTMP5 = s1 {DC00}
        ;                                    \
        VPADDD  XTMP0, XTMP5, XDWORD0;       \ // XDWORD0 = {W[3], W[2], W[1], W[0]}
        XORL    T1, y1;                      \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2)       // S0
@@ -704,8 +696,7 @@ avx2_loop0: // at each iteration works with one block (512 bit)
        VMOVDQU (2*32)(INP), XTMP2
        VMOVDQU (3*32)(INP), XTMP3
 
-       MOVQ    $flip_mask<>(SB), BP // BYTE_FLIP_MASK
-       VMOVDQU (BP), BYTE_FLIP_MASK
+       VMOVDQU flip_mask<>(SB), BYTE_FLIP_MASK
 
        // Apply Byte Flip Mask: LE -> BE
        VPSHUFB BYTE_FLIP_MASK, XTMP0, XTMP0
@@ -843,8 +834,7 @@ avx2_do_last_block:
        VMOVDQU 32(INP), XWORD2
        VMOVDQU 48(INP), XWORD3
 
-       MOVQ    $flip_mask<>(SB), BP
-       VMOVDQU (BP), X_BYTE_FLIP_MASK
+       VMOVDQU flip_mask<>(SB), BYTE_FLIP_MASK
 
        VPSHUFB X_BYTE_FLIP_MASK, XWORD0, XWORD0
        VPSHUFB X_BYTE_FLIP_MASK, XWORD1, XWORD1

diff --git a/src/crypto/sha512/sha512block_amd64.s b/src/crypto/sha512/sha512block_amd64.s
index 5b4242061537a48951c808fbc13e6c11a3080ace..a02356607ec9a55d758c3e639e784387daa43ef9 100644 (file)
--- a/src/crypto/sha512/sha512block_amd64.s
+++ b/src/crypto/sha512/sha512block_amd64.s
@@ -340,8 +340,7 @@ TEXT ·blockAVX2(SB), NOSPLIT, $56-32
        MOVQ (6*8)(SI), R10
        MOVQ (7*8)(SI), R11
 
-       MOVQ    $PSHUFFLE_BYTE_FLIP_MASK<>(SB), R12
-       VMOVDQU (R12), Y9
+       VMOVDQU PSHUFFLE_BYTE_FLIP_MASK<>(SB), Y9
 
 loop0:
        MOVQ ·_K+0(SB), BP
@@ -419,9 +418,7 @@ loop1:
 
        VPERM2F128 $0x0, Y0, Y0, Y4
 
-       MOVQ $MASK_YMM_LO<>(SB), R13
-
-       VPAND (R13), Y0, Y0
+       VPAND MASK_YMM_LO<>(SB), Y0, Y0
 
        VPERM2F128 $0x11, Y7, Y7, Y2
        VPSRLQ     $6, Y2, Y8
@@ -620,8 +617,7 @@ loop1:
 
        VPERM2F128 $0x0, Y0, Y0, Y5
 
-       MOVQ  $MASK_YMM_LO<>(SB), R13
-       VPAND (R13), Y0, Y0
+       VPAND MASK_YMM_LO<>(SB), Y0, Y0
 
        VPERM2F128 $0x11, Y4, Y4, Y2
        VPSRLQ     $6, Y2, Y8
@@ -820,8 +816,7 @@ loop1:
 
        VPERM2F128 $0x0, Y0, Y0, Y6
 
-       MOVQ  $MASK_YMM_LO<>(SB), R13
-       VPAND (R13), Y0, Y0
+       VPAND MASK_YMM_LO<>(SB), Y0, Y0
 
        VPERM2F128 $0x11, Y5, Y5, Y2
        VPSRLQ     $6, Y2, Y8
@@ -1021,8 +1016,7 @@ loop1:
 
        VPERM2F128 $0x0, Y0, Y0, Y7
 
-       MOVQ  $MASK_YMM_LO<>(SB), R13
-       VPAND (R13), Y0, Y0
+       VPAND MASK_YMM_LO<>(SB), Y0, Y0
 
        VPERM2F128 $0x11, Y6, Y6, Y2
        VPSRLQ     $6, Y2, Y8