wolfcrypt/src/wc_mlkem_poly.c and configure.ac: add support for WC_MLKEM_NO_ASM, and add gates to support WC_SHA3_NO_ASM;

wolfcrypt/src/sha3.c and wolfssl/wolfcrypt/sha3.h: BlockSha3() now always WOLFSSL_LOCAL (never static) to support calls from MLKEM implementation.

Author: douzzer

configure.ac

@@ -1502,6 +1502,9 @@ do
   ml-kem)
     ENABLED_ML_KEM=yes
     ;;
+  noasm)
+    AM_CFLAGS="$AM_CFLAGS -DWC_MLKEM_NO_ASM"
+    ;;
   *)
     AC_MSG_ERROR([Invalid choice for MLKEM []: $ENABLED_MLKEM.])
     break;;

wolfcrypt/src/sha3.c

@@ -248,9 +248,6 @@ while (0)
  *
  * s  The state.
  */
-#ifndef USE_INTEL_SPEEDUP
-static
-#endif
 void BlockSha3(word64* s)
 {
     byte i, x, y;
@@ -541,9 +538,6 @@ while (0)
  *
  * s  The state.
  */
-#ifndef USE_INTEL_SPEEDUP
-static
-#endif
 void BlockSha3(word64* s)
 {
     word64 n[25];

wolfcrypt/src/wc_mlkem_poly.c

@@ -69,6 +69,12 @@
 
 #include <wolfssl/wolfcrypt/libwolfssl_sources.h>
 
+#ifdef WC_MLKEM_NO_ASM
+    #undef USE_INTEL_SPEEDUP
+    #undef WOLFSSL_ARMASM
+    #undef WOLFSSL_RISCV_ASM
+#endif
+
 #include <wolfssl/wolfcrypt/wc_mlkem.h>
 #include <wolfssl/wolfcrypt/cpuid.h>
 
@@ -2481,6 +2487,7 @@ static int mlkem_gen_matrix_k3_avx2(sword16* a, byte* seed, int transposed)
     XMEMSET(state + 5, 0, sizeof(*state) * (25 - 5));
     state[20] = W64LIT(0x8000000000000000);
     for (i = 0; i < GEN_MATRIX_SIZE; i += SHA3_128_BYTES) {
+#ifndef WC_SHA3_NO_ASM
         if (IS_INTEL_BMI2(cpuid_flags)) {
             sha3_block_bmi2(state);
         }
@@ -2489,13 +2496,16 @@ static int mlkem_gen_matrix_k3_avx2(sword16* a, byte* seed, int transposed)
             sha3_block_avx2(state);
             RESTORE_VECTOR_REGISTERS();
         }
-        else {
+        else
+#endif /* !WC_SHA3_NO_ASM */
+        {
             BlockSha3(state);
         }
         XMEMCPY(rand + i, state, SHA3_128_BYTES);
     }
     ctr0 = mlkem_rej_uniform_n_avx2(a, MLKEM_N, rand, GEN_MATRIX_SIZE);
     while (ctr0 < MLKEM_N) {
+#ifndef WC_SHA3_NO_ASM
         if (IS_INTEL_BMI2(cpuid_flags)) {
             sha3_block_bmi2(state);
         }
@@ -2504,7 +2514,9 @@ static int mlkem_gen_matrix_k3_avx2(sword16* a, byte* seed, int transposed)
             sha3_block_avx2(state);
             RESTORE_VECTOR_REGISTERS();
         }
-        else {
+        else
+#endif /* !WC_SHA3_NO_ASM */
+        {
             BlockSha3(state);
         }
         XMEMCPY(rand, state, SHA3_128_BYTES);
@@ -3054,6 +3066,7 @@ static int mlkem_prf(wc_Shake* shake256, byte* out, unsigned int outLen,
         unsigned int len = min(outLen, WC_SHA3_256_BLOCK_SIZE);
 
         /* Perform a block operation on the state for next block of output. */
+#ifndef WC_SHA3_NO_ASM
         if (IS_INTEL_BMI2(cpuid_flags)) {
             sha3_block_bmi2(state);
         }
@@ -3062,7 +3075,9 @@ static int mlkem_prf(wc_Shake* shake256, byte* out, unsigned int outLen,
             sha3_block_avx2(state);
             RESTORE_VECTOR_REGISTERS();
         }
-        else {
+        else
+#endif /* !WC_SHA3_NO_ASM */
+        {
             BlockSha3(state);
         }
 
@@ -3109,14 +3124,17 @@ int mlkem_kdf(byte* seed, int seedLen, byte* out, int outLen)
     XMEMSET(state + len64 + 1, 0, (25 - len64 - 1) * sizeof(word64));
     state[WC_SHA3_256_COUNT - 1] = W64LIT(0x8000000000000000);
 
+#ifndef WC_SHA3_NO_ASM
     if (IS_INTEL_BMI2(cpuid_flags)) {
         sha3_block_bmi2(state);
     }
     else if (IS_INTEL_AVX2(cpuid_flags) && (SAVE_VECTOR_REGISTERS2() == 0)) {
         sha3_block_avx2(state);
         RESTORE_VECTOR_REGISTERS();
     }
-    else {
+    else
+#endif
+    {
         BlockSha3(state);
     }
     XMEMCPY(out, state, outLen);
@@ -4121,14 +4139,17 @@ static int mlkem_get_noise_eta2_avx2(MLKEM_PRF_T* prf, sword16* p,
     state[WC_SHA3_256_COUNT - 1] = W64LIT(0x8000000000000000);
 
     /* Perform a block operation on the state for next block of output. */
+#ifndef WC_SHA3_NO_ASM
     if (IS_INTEL_BMI2(cpuid_flags)) {
         sha3_block_bmi2(state);
     }
     else if (IS_INTEL_AVX2(cpuid_flags) && (SAVE_VECTOR_REGISTERS2() == 0)) {
         sha3_block_avx2(state);
         RESTORE_VECTOR_REGISTERS();
     }
-    else {
+    else
+#endif /* !WC_SHA3_NO_ASM */
+    {
         BlockSha3(state);
     }
     mlkem_cbd_eta2_avx2(p, (byte*)state);

wolfssl/wolfcrypt/sha3.h

@@ -220,8 +220,14 @@ WOLFSSL_API int wc_Shake256_Copy(wc_Shake* src, wc_Sha3* dst);
     WOLFSSL_API int wc_Sha3_GetFlags(wc_Sha3* sha3, word32* flags);
 #endif
 
+WOLFSSL_LOCAL void BlockSha3(word64 *s);
+
 #ifdef WC_SHA3_NO_ASM
 /* asm speedups disabled */
+#if defined(USE_INTEL_SPEEDUP) && !defined(WC_MLKEM_NO_ASM)
+    /* native ML-KEM uses this directly. */
+    WOLFSSL_LOCAL void sha3_blocksx4_avx2(word64* s);
+#endif
 #elif defined(USE_INTEL_SPEEDUP)
 WOLFSSL_LOCAL void sha3_block_n_bmi2(word64* s, const byte* data, word32 n,
     word64 c);