From 714f75d55187e15dec9c51577cb921af48033d31 Mon Sep 17 00:00:00 2001 From: Jens Steube Date: Fri, 5 Feb 2016 17:43:31 +0100 Subject: [PATCH] Converted to new SIMD: -m 1460 -a 0 --- OpenCL/m01460_a0.cl | 172 +++++++++++++++----------------------------- 1 file changed, 58 insertions(+), 114 deletions(-) diff --git a/OpenCL/m01460_a0.cl b/OpenCL/m01460_a0.cl index 2737cf409..d75069127 100644 --- a/OpenCL/m01460_a0.cl +++ b/OpenCL/m01460_a0.cl @@ -5,6 +5,8 @@ #define _SHA256_ +#define NEW_SIMD_CODE + #include "include/constants.h" #include "include/kernel_vendor.h" @@ -18,9 +20,7 @@ #include "OpenCL/common.c" #include "include/rp_kernel.h" #include "OpenCL/rp.c" - -#define COMPARE_S "OpenCL/check_single_comp4.c" -#define COMPARE_M "OpenCL/check_multi_comp4.c" +#include "OpenCL/simd.c" __constant u32 k_sha256[64] = { @@ -42,33 +42,33 @@ __constant u32 k_sha256[64] = SHA256C3c, SHA256C3d, SHA256C3e, SHA256C3f, }; -static void sha256_transform (const u32 w0[4], const u32 w1[4], const u32 w2[4], const u32 w3[4], u32 digest[8]) +static void sha256_transform (const u32x w0[4], const u32x w1[4], const u32x w2[4], const u32x w3[4], u32x digest[8]) { - u32 a = digest[0]; - u32 b = digest[1]; - u32 c = digest[2]; - u32 d = digest[3]; - u32 e = digest[4]; - u32 f = digest[5]; - u32 g = digest[6]; - u32 h = digest[7]; + u32x a = digest[0]; + u32x b = digest[1]; + u32x c = digest[2]; + u32x d = digest[3]; + u32x e = digest[4]; + u32x f = digest[5]; + u32x g = digest[6]; + u32x h = digest[7]; - u32 w0_t = w0[0]; - u32 w1_t = w0[1]; - u32 w2_t = w0[2]; - u32 w3_t = w0[3]; - u32 w4_t = w1[0]; - u32 w5_t = w1[1]; - u32 w6_t = w1[2]; - u32 w7_t = w1[3]; - u32 w8_t = w2[0]; - u32 w9_t = w2[1]; - u32 wa_t = w2[2]; - u32 wb_t = w2[3]; - u32 wc_t = w3[0]; - u32 wd_t = w3[1]; - u32 we_t = w3[2]; - u32 wf_t = w3[3]; + u32x w0_t = w0[0]; + u32x w1_t = w0[1]; + u32x w2_t = w0[2]; + u32x w3_t = w0[3]; + u32x w4_t = w1[0]; + u32x w5_t = w1[1]; + u32x w6_t = w1[2]; + u32x w7_t = w1[3]; + u32x w8_t = w2[0]; + u32x w9_t = w2[1]; + u32x wa_t = w2[2]; + u32x wb_t = w2[3]; + u32x wc_t = w3[0]; + u32x wd_t = w3[1]; + u32x we_t = w3[2]; + u32x wf_t = w3[3]; #define ROUND_EXPAND() \ { \ @@ -128,7 +128,7 @@ static void sha256_transform (const u32 w0[4], const u32 w1[4], const u32 w2[4], digest[7] += h; } -static void hmac_sha256_pad (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], u32 ipad[8], u32 opad[8]) +static void hmac_sha256_pad (u32x w0[4], u32x w1[4], u32x w2[4], u32x w3[4], u32x ipad[8], u32x opad[8]) { w0[0] = w0[0] ^ 0x36363636; w0[1] = w0[1] ^ 0x36363636; @@ -187,7 +187,7 @@ static void hmac_sha256_pad (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], u32 ipa sha256_transform (w0, w1, w2, w3, opad); } -static void hmac_sha256_run (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], u32 ipad[8], u32 opad[8], u32 digest[8]) +static void hmac_sha256_run (u32x w0[4], u32x w1[4], u32x w2[4], u32x w3[4], u32x ipad[8], u32x opad[8], u32x digest[8]) { digest[0] = ipad[0]; digest[1] = ipad[1]; @@ -283,36 +283,36 @@ __kernel void m01460_m04 (__global pw_t *pws, __global kernel_rule_t * rules_bu * pads */ - u32 w0_t[4]; + u32x w0_t[4]; w0_t[0] = swap32 (salt_buf0[0]); w0_t[1] = swap32 (salt_buf0[1]); w0_t[2] = swap32 (salt_buf0[2]); w0_t[3] = swap32 (salt_buf0[3]); - u32 w1_t[4]; + u32x w1_t[4]; w1_t[0] = swap32 (salt_buf1[0]); w1_t[1] = swap32 (salt_buf1[1]); w1_t[2] = swap32 (salt_buf1[2]); w1_t[3] = swap32 (salt_buf1[3]); - u32 w2_t[4]; + u32x w2_t[4]; w2_t[0] = 0; w2_t[1] = 0; w2_t[2] = 0; w2_t[3] = 0; - u32 w3_t[4]; + u32x w3_t[4]; w3_t[0] = 0; w3_t[1] = 0; w3_t[2] = 0; w3_t[3] = 0; - u32 ipad[8]; - u32 opad[8]; + u32x ipad[8]; + u32x opad[8]; hmac_sha256_pad (w0_t, w1_t, w2_t, w3_t, ipad, opad); @@ -320,37 +320,14 @@ __kernel void m01460_m04 (__global pw_t *pws, __global kernel_rule_t * rules_bu * loop */ - for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < rules_cnt; il_pos += VECT_SIZE) { - u32 w0[4]; + u32x w0[4] = { 0 }; + u32x w1[4] = { 0 }; + u32x w2[4] = { 0 }; + u32x w3[4] = { 0 }; - w0[0] = pw_buf0[0]; - w0[1] = pw_buf0[1]; - w0[2] = pw_buf0[2]; - w0[3] = pw_buf0[3]; - - u32 w1[4]; - - w1[0] = pw_buf1[0]; - w1[1] = pw_buf1[1]; - w1[2] = pw_buf1[2]; - w1[3] = pw_buf1[3]; - - u32 w2[4]; - - w2[0] = 0; - w2[1] = 0; - w2[2] = 0; - w2[3] = 0; - - u32 w3[4]; - - w3[0] = 0; - w3[1] = 0; - w3[2] = 0; - w3[3] = 0; - - const u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len); + const u32 out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); append_0x80_2x4 (w0, w1, out_len); @@ -371,16 +348,11 @@ __kernel void m01460_m04 (__global pw_t *pws, __global kernel_rule_t * rules_bu w3_t[2] = 0; w3_t[3] = (64 + out_len) * 8; - u32 digest[8]; + u32x digest[8]; hmac_sha256_run (w0_t, w1_t, w2_t, w3_t, ipad, opad, digest); - const u32 r0 = digest[3]; - const u32 r1 = digest[7]; - const u32 r2 = digest[2]; - const u32 r3 = digest[6]; - - #include COMPARE_M + COMPARE_M_SIMD (digest[3], digest[7], digest[2], digest[6]); } } @@ -446,36 +418,36 @@ __kernel void m01460_s04 (__global pw_t *pws, __global kernel_rule_t * rules_bu * pads */ - u32 w0_t[4]; + u32x w0_t[4]; w0_t[0] = swap32 (salt_buf0[0]); w0_t[1] = swap32 (salt_buf0[1]); w0_t[2] = swap32 (salt_buf0[2]); w0_t[3] = swap32 (salt_buf0[3]); - u32 w1_t[4]; + u32x w1_t[4]; w1_t[0] = swap32 (salt_buf1[0]); w1_t[1] = swap32 (salt_buf1[1]); w1_t[2] = swap32 (salt_buf1[2]); w1_t[3] = swap32 (salt_buf1[3]); - u32 w2_t[4]; + u32x w2_t[4]; w2_t[0] = 0; w2_t[1] = 0; w2_t[2] = 0; w2_t[3] = 0; - u32 w3_t[4]; + u32x w3_t[4]; w3_t[0] = 0; w3_t[1] = 0; w3_t[2] = 0; w3_t[3] = 0; - u32 ipad[8]; - u32 opad[8]; + u32x ipad[8]; + u32x opad[8]; hmac_sha256_pad (w0_t, w1_t, w2_t, w3_t, ipad, opad); @@ -495,37 +467,14 @@ __kernel void m01460_s04 (__global pw_t *pws, __global kernel_rule_t * rules_bu * loop */ - for (u32 il_pos = 0; il_pos < rules_cnt; il_pos++) + for (u32 il_pos = 0; il_pos < rules_cnt; il_pos += VECT_SIZE) { - u32 w0[4]; + u32x w0[4] = { 0 }; + u32x w1[4] = { 0 }; + u32x w2[4] = { 0 }; + u32x w3[4] = { 0 }; - w0[0] = pw_buf0[0]; - w0[1] = pw_buf0[1]; - w0[2] = pw_buf0[2]; - w0[3] = pw_buf0[3]; - - u32 w1[4]; - - w1[0] = pw_buf1[0]; - w1[1] = pw_buf1[1]; - w1[2] = pw_buf1[2]; - w1[3] = pw_buf1[3]; - - u32 w2[4]; - - w2[0] = 0; - w2[1] = 0; - w2[2] = 0; - w2[3] = 0; - - u32 w3[4]; - - w3[0] = 0; - w3[1] = 0; - w3[2] = 0; - w3[3] = 0; - - const u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len); + const u32 out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1); append_0x80_2x4 (w0, w1, out_len); @@ -546,16 +495,11 @@ __kernel void m01460_s04 (__global pw_t *pws, __global kernel_rule_t * rules_bu w3_t[2] = 0; w3_t[3] = (64 + out_len) * 8; - u32 digest[8]; + u32x digest[8]; hmac_sha256_run (w0_t, w1_t, w2_t, w3_t, ipad, opad, digest); - const u32 r0 = digest[3]; - const u32 r1 = digest[7]; - const u32 r2 = digest[2]; - const u32 r3 = digest[6]; - - #include COMPARE_S + COMPARE_S_SIMD (digest[3], digest[7], digest[2], digest[6]); } }