crypto: arm/ghash-ce - implement support for 4-way aggregation
Speed up the GHASH algorithm based on 64-bit polynomial multiplication
by adding support for 4-way aggregation. This improves throughput by
~85% on Cortex-A53, from 1.7 cycles per byte to 0.9 cycles per byte.
When combined with AES into GCM, throughput improves by ~25%, from
3.8 cycles per byte to 3.0 cycles per byte.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
diff --git a/arch/arm/crypto/ghash-ce-core.S b/arch/arm/crypto/ghash-ce-core.S
index 2f78c10..406009a 100644
--- a/arch/arm/crypto/ghash-ce-core.S
+++ b/arch/arm/crypto/ghash-ce-core.S
@@ -63,6 +63,33 @@
k48 .req d31
SHASH2_p64 .req d31
+ HH .req q10
+ HH3 .req q11
+ HH4 .req q12
+ HH34 .req q13
+
+ HH_L .req d20
+ HH_H .req d21
+ HH3_L .req d22
+ HH3_H .req d23
+ HH4_L .req d24
+ HH4_H .req d25
+ HH34_L .req d26
+ HH34_H .req d27
+ SHASH2_H .req d29
+
+ XL2 .req q5
+ XM2 .req q6
+ XH2 .req q7
+ T3 .req q8
+
+ XL2_L .req d10
+ XL2_H .req d11
+ XM2_L .req d12
+ XM2_H .req d13
+ T3_L .req d16
+ T3_H .req d17
+
.text
.fpu crypto-neon-fp-armv8
@@ -175,12 +202,77 @@
beq 0f
vld1.64 {T1}, [ip]
teq r0, #0
- b 1f
+ b 3f
-0: vld1.64 {T1}, [r2]!
+0: .ifc \pn, p64
+ tst r0, #3 // skip until #blocks is a
+ bne 2f // round multiple of 4
+
+ vld1.8 {XL2-XM2}, [r2]!
+1: vld1.8 {T3-T2}, [r2]!
+ vrev64.8 XL2, XL2
+ vrev64.8 XM2, XM2
+
+ subs r0, r0, #4
+
+ vext.8 T1, XL2, XL2, #8
+ veor XL2_H, XL2_H, XL_L
+ veor XL, XL, T1
+
+ vrev64.8 T3, T3
+ vrev64.8 T1, T2
+
+ vmull.p64 XH, HH4_H, XL_H // a1 * b1
+ veor XL2_H, XL2_H, XL_H
+ vmull.p64 XL, HH4_L, XL_L // a0 * b0
+ vmull.p64 XM, HH34_H, XL2_H // (a1 + a0)(b1 + b0)
+
+ vmull.p64 XH2, HH3_H, XM2_L // a1 * b1
+ veor XM2_L, XM2_L, XM2_H
+ vmull.p64 XL2, HH3_L, XM2_H // a0 * b0
+ vmull.p64 XM2, HH34_L, XM2_L // (a1 + a0)(b1 + b0)
+
+ veor XH, XH, XH2
+ veor XL, XL, XL2
+ veor XM, XM, XM2
+
+ vmull.p64 XH2, HH_H, T3_L // a1 * b1
+ veor T3_L, T3_L, T3_H
+ vmull.p64 XL2, HH_L, T3_H // a0 * b0
+ vmull.p64 XM2, SHASH2_H, T3_L // (a1 + a0)(b1 + b0)
+
+ veor XH, XH, XH2
+ veor XL, XL, XL2
+ veor XM, XM, XM2
+
+ vmull.p64 XH2, SHASH_H, T1_L // a1 * b1
+ veor T1_L, T1_L, T1_H
+ vmull.p64 XL2, SHASH_L, T1_H // a0 * b0
+ vmull.p64 XM2, SHASH2_p64, T1_L // (a1 + a0)(b1 + b0)
+
+ veor XH, XH, XH2
+ veor XL, XL, XL2
+ veor XM, XM, XM2
+
+ beq 4f
+
+ vld1.8 {XL2-XM2}, [r2]!
+
+ veor T1, XL, XH
+ veor XM, XM, T1
+
+ __pmull_reduce_p64
+
+ veor T1, T1, XH
+ veor XL, XL, T1
+
+ b 1b
+ .endif
+
+2: vld1.64 {T1}, [r2]!
subs r0, r0, #1
-1: /* multiply XL by SHASH in GF(2^128) */
+3: /* multiply XL by SHASH in GF(2^128) */
#ifndef CONFIG_CPU_BIG_ENDIAN
vrev64.8 T1, T1
#endif
@@ -193,7 +285,7 @@
__pmull_\pn XL, XL_L, SHASH_L, s1l, s2l, s3l, s4l @ a0 * b0
__pmull_\pn XM, T1_L, SHASH2_\pn @ (a1+a0)(b1+b0)
- veor T1, XL, XH
+4: veor T1, XL, XH
veor XM, XM, T1
__pmull_reduce_\pn
@@ -212,8 +304,14 @@
* struct ghash_key const *k, const char *head)
*/
ENTRY(pmull_ghash_update_p64)
- vld1.64 {SHASH}, [r3]
+ vld1.64 {SHASH}, [r3]!
+ vld1.64 {HH}, [r3]!
+ vld1.64 {HH3-HH4}, [r3]
+
veor SHASH2_p64, SHASH_L, SHASH_H
+ veor SHASH2_H, HH_L, HH_H
+ veor HH34_L, HH3_L, HH3_H
+ veor HH34_H, HH4_L, HH4_H
vmov.i8 MASK, #0xe1
vshl.u64 MASK, MASK, #57