OpenCL GOST/Streebog: Request not to inline update/final functions

For the corresponding crypt formats, build times went from minutes (worst
case) to a few seconds while speeds were unaffected and gost94crypt-opencl
stopped failing self-test on super's AMD Vega.

Author: magnumripper

run/opencl/gost12512hash_kernel.cl

@@ -12,11 +12,6 @@
 #define STREEBOG_LOCAL_AX       1
 #endif
 #define STREEBOG_VECTOR         1
-/*
- * Without unroll: Build time: 45.125 s, binary size 2542391, 10402 c/s
- * Pragma unroll:  Build time: 21 min 14.788 s, binary size 11502092, 10051 c/s
- * Manual unroll:  Build time: 4 min 26.142 s, binary size 5961490, 9959 c/s
- */
 #define STREEBOG_UNROLL         0
 #include "opencl_streebog.h"
 

run/opencl/opencl_gost94.h

@@ -300,7 +300,7 @@ inline void rhash_gost94_compute_sum_and_hash(gost94_ctx * ctx, const uint* bloc
  * @param msg message chunk
  * @param size length of the message chunk
  */
-inline void gost94_update(gost94_ctx *ctx, const uchar* msg, uint size, MAYBE_LOCAL const rhash_gost94_sbox *sbox)
+__attribute__((noinline)) void gost94_update(gost94_ctx *ctx, const uchar* msg, uint size, MAYBE_LOCAL const rhash_gost94_sbox *sbox)
 {
 	uint index = ctx->length & 31;
 	ctx->length += size;
@@ -361,7 +361,7 @@ inline void rhash_u32_swap_copy(void* to, const void* from, uint length) {
  * @param ctx the algorithm context containing current hashing state
  * @param result calculated hash in binary form
  */
-inline void gost94_final(gost94_ctx *ctx, uchar *result, MAYBE_LOCAL const rhash_gost94_sbox *sbox)
+__attribute__((noinline)) void gost94_final(gost94_ctx *ctx, uchar *result, MAYBE_LOCAL const rhash_gost94_sbox *sbox)
 {
 	uint  index = ctx->length & 31;
 	uint* msg32 = (uint*)ctx->message;

run/opencl/opencl_streebog.h

@@ -906,7 +906,7 @@ stage3(GOST34112012Context *CTX, __local localbuf *loc_buf)
 	g0(&(CTX->h), &(CTX->Sigma), loc_buf);
 }
 
-inline void
+__attribute__((noinline)) void
 GOST34112012Update(GOST34112012Context *CTX, const uchar *data, uint len, __local localbuf *loc_buf)
 {
 	if (CTX->bufsize) {
@@ -937,7 +937,7 @@ GOST34112012Update(GOST34112012Context *CTX, const uchar *data, uint len, __loca
 	}
 }
 
-inline void
+__attribute__((noinline)) void
 GOST34112012Final(GOST34112012Context *CTX,
 #if STREEBOG512CRYPT
                   uint512_u