pq_crypto 0.6.1 → 0.6.2

data/ext/pqcrypto/vendor/mldsa-native/mldsa/src/fips202/native/armv81m/src/state_xor_bytes_x4_mve.S

@@ -0,0 +1,355 @@
+/*
+ * Copyright (c) The mlkem-native project authors
+ * Copyright (c) The mldsa-native project authors
+ * Copyright (c) 2026 Arm Limited
+ * SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT
+ */
+
+// ---------------------------------------------------------------------------
+// Overview
+// ---------------------------------------------------------------------------
+// MVE/Helium implementation of KeccakF1600x4_StateXORBytes.
+//
+// void KeccakF1600x4_StateXORBytes(state, d0, d1, d2, d3, offset, length)
+//
+// Reads 'length' plain bytes from each of four input buffers (d0..d3),
+// splits every byte into its even and odd bits (bit-interleaving), and
+// XORs the result into the Keccak state starting at byte 'offset'.
+//
+// ---------------------------------------------------------------------------
+// Bit-interleaving background
+// ---------------------------------------------------------------------------
+// Each 64-bit Keccak lane is stored as two 32-bit words:
+//   even half -- bits 0, 2, 4, ..., 62 of the lane
+//   odd half  -- bits 1, 3, 5, ..., 63 of the lane
+// This representation allows 64-bit lane rotations (used in the Keccak
+// round function) to be implemented as pairs of 32-bit rotations.
+//
+// Batched (x4) processing:
+//   Four Keccak instances are processed as a batch.  Their states are
+//   stored interleaved in a single 800-byte buffer: first the even
+//   halves of all 25 lanes (400 bytes), then the odd halves (400 bytes).
+//   Within each 16-byte row, the four u32 words correspond to
+//   instances 0..3 of the same lane, enabling SIMD-parallel operations
+//   across all four instances.
+//
+// State memory layout (25 lanes x 4 instances x 2 halves):
+//   S[i][l]_even/odd = even/odd half of lane l, instance i  (u32)
+//   Each row is 16 bytes (one Q-register).
+//   Offset  Contents
+//     0     S[0][ 0]_even, S[1][ 0]_even, S[2][ 0]_even, S[3][ 0]_even
+//    16     S[0][ 1]_even, S[1][ 1]_even, S[2][ 1]_even, S[3][ 1]_even
+//    ...
+//   384     S[0][24]_even, S[1][24]_even, S[2][24]_even, S[3][24]_even
+//   400     S[0][ 0]_odd,  S[1][ 0]_odd,  S[2][ 0]_odd,  S[3][ 0]_odd
+//   416     S[0][ 1]_odd,  S[1][ 1]_odd,  S[2][ 1]_odd,  S[3][ 1]_odd
+//    ...
+//   784     S[0][24]_odd,  S[1][24]_odd,  S[2][24]_odd,  S[3][24]_odd
+//
+// ---------------------------------------------------------------------------
+// Three-phase structure
+// ---------------------------------------------------------------------------
+//   Prologue -- if offset is not 8-byte aligned, absorb
+//               min(length, 8-(offset%8)) bytes via predicated byte loads.
+//   Main     -- process full 8-byte groups via word-level gather loads,
+//               bit-interleave, then VEOR into even/odd state halves.
+//   Tail     -- absorb remaining <8 bytes via predicated byte loads.
+
+#include "../../../../common.h"
+#if defined(MLD_FIPS202_ARMV81M_NEED_X4) && \
+    !defined(MLD_CONFIG_MULTILEVEL_NO_SHARED)
+
+/*
+ * WARNING: This file is auto-derived from the mldsa-native source file
+ *   dev/fips202/armv81m/src/state_xor_bytes_x4_mve.S using scripts/simpasm. Do not modify it directly.
+ */
+
+.thumb
+.syntax unified
+
+.text
+.balign 4
+.global MLD_ASM_NAMESPACE(keccak_f1600_x4_state_xor_bytes_asm)
+MLD_ASM_FN_SYMBOL(keccak_f1600_x4_state_xor_bytes_asm)
+
+        .cfi_startproc
+        push.w {r4, r5, r6, r7, r8, r9, r10, r11, r12, lr}
+        .cfi_adjust_cfa_offset 0x28
+        .cfi_rel_offset r4, 0x0
+        .cfi_rel_offset r5, 0x4
+        .cfi_rel_offset r6, 0x8
+        .cfi_rel_offset r7, 0xc
+        .cfi_rel_offset r8, 0x10
+        .cfi_rel_offset r9, 0x14
+        .cfi_rel_offset r10, 0x18
+        .cfi_rel_offset r11, 0x1c
+        .cfi_rel_offset lr, 0x24
+        vpush {d8, d9, d10, d11, d12, d13, d14, d15}
+        .cfi_adjust_cfa_offset 0x40
+        .cfi_rel_offset d8, 0x0
+        .cfi_rel_offset d9, 0x8
+        .cfi_rel_offset d10, 0x10
+        .cfi_rel_offset d11, 0x18
+        .cfi_rel_offset d12, 0x20
+        .cfi_rel_offset d13, 0x28
+        .cfi_rel_offset d14, 0x30
+        .cfi_rel_offset d15, 0x38
+        ldr r4, [sp, #0x68]
+        ldr.w r10, [sp, #0x6c]
+        ldr r6, [sp, #0x70]
+        cmp r6, #0x0
+        beq.w Lkeccak_f1600_x4_state_xor_bytes_asm_exit @ imm = #0x346
+        and r5, r10, #0x7
+        bic r9, r10, #0x7
+        add.w r8, r0, r9, lsl #1
+        add.w r7, r8, #0x190
+        cmp r5, #0x0
+        beq.w Lkeccak_f1600_x4_state_xor_bytes_asm_pre_main @ imm = #0x12c
+        subs r1, r1, r5
+        subs r2, r2, r5
+        subs r3, r3, r5
+        subs r4, r4, r5
+        rsb.w lr, r5, #0x8
+        cmp r6, lr
+        it ls
+        movls lr, r6
+        subs.w r6, r6, lr
+        vctp.8 lr
+        vmrs r11, p0
+        lsl.w r11, r11, r5
+        vmsr p0, r11
+        vpstttt
+        vldrbt.u8 q0, [r1], #4
+        vldrbt.u8 q1, [r2], #4
+        vldrbt.u8 q2, [r3], #4
+        vldrbt.u8 q3, [r4], #4
+        vmov.f64 d1, d4
+        vmov.f64 d3, d6
+        vrev64.32 q2, q0
+        vrev64.32 q3, q1
+        movw r0, #0xf0f
+        vmsr p0, r0
+        vpsel q0, q0, q3
+        vpsel q1, q2, q1
+        vmov q2, q0
+        vmov q3, q1
+        vshr.u8 q4, q0, #0x2
+        vsli.8 q0, q4, #0x1
+        vshr.u8 q4, q0, #0x3
+        vsli.8 q0, q4, #0x2
+        vshr.u8 q4, q0, #0x4
+        vsli.8 q0, q4, #0x3
+        vshr.u16 q4, q0, #0x8
+        vsli.8 q0, q4, #0x4
+        vshr.u32 q4, q0, #0x10
+        vsli.16 q0, q4, #0x8
+        vshr.u8 q4, q3, #0x2
+        vsli.8 q3, q4, #0x1
+        vshr.u8 q4, q3, #0x3
+        vsli.8 q3, q4, #0x2
+        vshr.u8 q4, q3, #0x4
+        vsli.8 q3, q4, #0x3
+        vshr.u16 q4, q3, #0x8
+        vsli.8 q3, q4, #0x4
+        vshr.u32 q4, q3, #0x10
+        vsli.16 q3, q4, #0x8
+        vsli.32 q0, q3, #0x10
+        vshl.i8 q4, q2, #0x2
+        vsri.8 q2, q4, #0x1
+        vshl.i8 q4, q2, #0x3
+        vsri.8 q2, q4, #0x2
+        vshl.i8 q4, q2, #0x4
+        vsri.8 q2, q4, #0x3
+        vshl.i16 q4, q2, #0x8
+        vsri.8 q2, q4, #0x4
+        vshl.i32 q4, q2, #0x10
+        vsri.16 q2, q4, #0x8
+        vshl.i8 q4, q1, #0x2
+        vsri.8 q1, q4, #0x1
+        vshl.i8 q4, q1, #0x3
+        vsri.8 q1, q4, #0x2
+        vshl.i8 q4, q1, #0x4
+        vsri.8 q1, q4, #0x3
+        vshl.i16 q4, q1, #0x8
+        vsri.8 q1, q4, #0x4
+        vshl.i32 q4, q1, #0x10
+        vsri.16 q1, q4, #0x8
+        vsri.32 q1, q2, #0x10
+        vldrw.u32 q4, [r8]
+        vldrw.u32 q5, [r7]
+        veor q4, q4, q0
+        veor q5, q5, q1
+        vstrw.32 q4, [r8], #16
+        vstrw.32 q5, [r7], #16
+        vmov q7[2], q7[0], r1, r3
+        vmov q7[3], q7[1], r2, r4
+        b Lkeccak_f1600_x4_state_xor_bytes_asm_main_body @ imm = #0xe
+
+Lkeccak_f1600_x4_state_xor_bytes_asm_pre_main:
+        vmov q7[2], q7[0], r1, r3
+        vmov q7[3], q7[1], r2, r4
+        mov.w r0, #0x4
+        vsub.i32 q7, q7, r0
+
+Lkeccak_f1600_x4_state_xor_bytes_asm_main_body:
+        lsr.w lr, r6, #0x3
+        wls lr, lr, Lkeccak_f1600_x4_state_xor_bytes_asm_main_loop_end @ imm = #0xd4
+
+Lkeccak_f1600_x4_state_xor_bytes_asm_main_loop_start:
+        vldrw.u32 q0, [q7, #4]!
+        vldrw.u32 q1, [q7, #4]!
+        vmov q2, q0
+        vmov q3, q1
+        vshr.u8 q4, q0, #0x2
+        vsli.8 q0, q4, #0x1
+        vshr.u8 q4, q0, #0x3
+        vsli.8 q0, q4, #0x2
+        vshr.u8 q4, q0, #0x4
+        vsli.8 q0, q4, #0x3
+        vshr.u16 q4, q0, #0x8
+        vsli.8 q0, q4, #0x4
+        vshr.u32 q4, q0, #0x10
+        vsli.16 q0, q4, #0x8
+        vshr.u8 q4, q3, #0x2
+        vsli.8 q3, q4, #0x1
+        vshr.u8 q4, q3, #0x3
+        vsli.8 q3, q4, #0x2
+        vshr.u8 q4, q3, #0x4
+        vsli.8 q3, q4, #0x3
+        vshr.u16 q4, q3, #0x8
+        vsli.8 q3, q4, #0x4
+        vshr.u32 q4, q3, #0x10
+        vsli.16 q3, q4, #0x8
+        vsli.32 q0, q3, #0x10
+        vshl.i8 q4, q2, #0x2
+        vsri.8 q2, q4, #0x1
+        vshl.i8 q4, q2, #0x3
+        vsri.8 q2, q4, #0x2
+        vshl.i8 q4, q2, #0x4
+        vsri.8 q2, q4, #0x3
+        vshl.i16 q4, q2, #0x8
+        vsri.8 q2, q4, #0x4
+        vshl.i32 q4, q2, #0x10
+        vsri.16 q2, q4, #0x8
+        vshl.i8 q4, q1, #0x2
+        vsri.8 q1, q4, #0x1
+        vshl.i8 q4, q1, #0x3
+        vsri.8 q1, q4, #0x2
+        vshl.i8 q4, q1, #0x4
+        vsri.8 q1, q4, #0x3
+        vshl.i16 q4, q1, #0x8
+        vsri.8 q1, q4, #0x4
+        vshl.i32 q4, q1, #0x10
+        vsri.16 q1, q4, #0x8
+        vsri.32 q1, q2, #0x10
+        vldrw.u32 q4, [r8]
+        vldrw.u32 q5, [r7]
+        veor q4, q4, q0
+        veor q5, q5, q1
+        vstrw.32 q4, [r8], #16
+        vstrw.32 q5, [r7], #16
+        le lr, Lkeccak_f1600_x4_state_xor_bytes_asm_main_loop_start @ imm = #-0xd4
+
+Lkeccak_f1600_x4_state_xor_bytes_asm_main_loop_end:
+        ands r6, r6, #0x7
+        beq.w Lkeccak_f1600_x4_state_xor_bytes_asm_exit @ imm = #0x110
+        mov.w r0, #0x4
+        vadd.i32 q7, q7, r0
+        vmov r1, r3, q7[2], q7[0]
+        vmov r2, r4, q7[3], q7[1]
+        vctp.8 r6
+        vpstttt
+        vldrbt.u8 q0, [r1]
+        vldrbt.u8 q1, [r2]
+        vldrbt.u8 q2, [r3]
+        vldrbt.u8 q3, [r4]
+        vmov.f64 d1, d4
+        vmov.f64 d3, d6
+        vrev64.32 q2, q0
+        vrev64.32 q3, q1
+        movw r0, #0xf0f
+        vmsr p0, r0
+        vpsel q0, q0, q3
+        vpsel q1, q2, q1
+        vmov q2, q0
+        vmov q3, q1
+        vshr.u8 q4, q0, #0x2
+        vsli.8 q0, q4, #0x1
+        vshr.u8 q4, q0, #0x3
+        vsli.8 q0, q4, #0x2
+        vshr.u8 q4, q0, #0x4
+        vsli.8 q0, q4, #0x3
+        vshr.u16 q4, q0, #0x8
+        vsli.8 q0, q4, #0x4
+        vshr.u32 q4, q0, #0x10
+        vsli.16 q0, q4, #0x8
+        vshr.u8 q4, q3, #0x2
+        vsli.8 q3, q4, #0x1
+        vshr.u8 q4, q3, #0x3
+        vsli.8 q3, q4, #0x2
+        vshr.u8 q4, q3, #0x4
+        vsli.8 q3, q4, #0x3
+        vshr.u16 q4, q3, #0x8
+        vsli.8 q3, q4, #0x4
+        vshr.u32 q4, q3, #0x10
+        vsli.16 q3, q4, #0x8
+        vsli.32 q0, q3, #0x10
+        vshl.i8 q4, q2, #0x2
+        vsri.8 q2, q4, #0x1
+        vshl.i8 q4, q2, #0x3
+        vsri.8 q2, q4, #0x2
+        vshl.i8 q4, q2, #0x4
+        vsri.8 q2, q4, #0x3
+        vshl.i16 q4, q2, #0x8
+        vsri.8 q2, q4, #0x4
+        vshl.i32 q4, q2, #0x10
+        vsri.16 q2, q4, #0x8
+        vshl.i8 q4, q1, #0x2
+        vsri.8 q1, q4, #0x1
+        vshl.i8 q4, q1, #0x3
+        vsri.8 q1, q4, #0x2
+        vshl.i8 q4, q1, #0x4
+        vsri.8 q1, q4, #0x3
+        vshl.i16 q4, q1, #0x8
+        vsri.8 q1, q4, #0x4
+        vshl.i32 q4, q1, #0x10
+        vsri.16 q1, q4, #0x8
+        vsri.32 q1, q2, #0x10
+        vldrw.u32 q4, [r8]
+        vldrw.u32 q5, [r7]
+        veor q4, q4, q0
+        veor q5, q5, q1
+        vstrw.32 q4, [r8], #16
+        vstrw.32 q5, [r7], #16
+
+Lkeccak_f1600_x4_state_xor_bytes_asm_exit:
+        vpop {d8, d9, d10, d11, d12, d13, d14, d15}
+        .cfi_restore d8
+        .cfi_restore d9
+        .cfi_restore d10
+        .cfi_restore d11
+        .cfi_restore d12
+        .cfi_restore d13
+        .cfi_restore d14
+        .cfi_restore d15
+        .cfi_adjust_cfa_offset -0x40
+        pop.w {r4, r5, r6, r7, r8, r9, r10, r11, r12, pc}
+        .cfi_restore r4
+        .cfi_restore r5
+        .cfi_restore r6
+        .cfi_restore r7
+        .cfi_restore r8
+        .cfi_restore r9
+        .cfi_restore r10
+        .cfi_restore r11
+        .cfi_restore lr
+        .cfi_adjust_cfa_offset -0x28
+        .cfi_endproc
+
+MLD_ASM_FN_SIZE(keccak_f1600_x4_state_xor_bytes_asm)
+
+#endif /* MLD_FIPS202_ARMV81M_NEED_X4 && !MLD_CONFIG_MULTILEVEL_NO_SHARED */
+
+#if defined(__ELF__)
+.section .note.GNU-stack,"",%progbits
+#endif

data/ext/pqcrypto/vendor/mldsa-native/mldsa/src/fips202/native/auto.h

@@ -16,9 +16,14 @@
 #include "aarch64/auto.h"
 #endif
 
-#if defined(MLD_SYS_X86_64) && defined(MLD_SYS_X86_64_AVX2)
-#include "x86_64/xkcp.h"
-#endif
+#if defined(MLD_SYS_X86_64) && defined(MLD_SYS_X86_64_AVX2) &&               \
+    (!defined(MLD_CONFIG_NO_KEYPAIR_API) ||                                  \
+     !defined(MLD_CONFIG_NO_SIGN_API) || !defined(MLD_CONFIG_REDUCE_RAM)) && \
+    !defined(MLD_CONFIG_SERIAL_FIPS202_ONLY)
+#include "x86_64/keccak_f1600_x4_avx2.h"
+#endif /* MLD_SYS_X86_64 && MLD_SYS_X86_64_AVX2 && (!MLD_CONFIG_NO_KEYPAIR_API \
+          || !MLD_CONFIG_NO_SIGN_API || !MLD_CONFIG_REDUCE_RAM) &&             \
+          !MLD_CONFIG_SERIAL_FIPS202_ONLY */
 
 /* We do not yet include the FIPS202 backend for Armv8.1-M+MVE by default
  * as it is still experimental and undergoing review. */

data/ext/pqcrypto/vendor/mldsa-native/mldsa/src/fips202/native/x86_64/{xkcp.h → keccak_f1600_x4_avx2.h}

@@ -4,18 +4,19 @@
  * SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT
  */
 
-#ifndef MLD_FIPS202_NATIVE_X86_64_XKCP_H
-#define MLD_FIPS202_NATIVE_X86_64_XKCP_H
+#ifndef MLD_FIPS202_NATIVE_X86_64_KECCAK_F1600_X4_AVX2_H
+#define MLD_FIPS202_NATIVE_X86_64_KECCAK_F1600_X4_AVX2_H
 
 #include "../../../common.h"
 
-#define MLD_FIPS202_X86_64_XKCP
+#define MLD_FIPS202_X86_64_NEED_X4_AVX2
+
+/* Part of backend API */
+#define MLD_USE_FIPS202_X4_NATIVE
 
 #if !defined(__ASSEMBLER__)
 #include "../api.h"
-#include "src/KeccakP_1600_times4_SIMD256.h"
-
-#define MLD_USE_FIPS202_X4_NATIVE
+#include "src/fips202_native_x86_64.h"
 MLD_MUST_CHECK_RETURN_VALUE
 static MLD_INLINE int mld_keccak_f1600_x4_native(uint64_t *state)
 {
@@ -23,9 +24,11 @@ static MLD_INLINE int mld_keccak_f1600_x4_native(uint64_t *state)
   {
     return MLD_NATIVE_FUNC_FALLBACK;
   }
-  mld_keccakf1600x4_permute24(state);
+
+  mld_keccak_f1600_x4_avx2_asm(state, mld_keccakf1600_round_constants,
+                               mld_keccak_rho8, mld_keccak_rho56);
   return MLD_NATIVE_FUNC_SUCCESS;
 }
 #endif /* !__ASSEMBLER__ */
 
-#endif /* !MLD_FIPS202_NATIVE_X86_64_XKCP_H */
+#endif /* !MLD_FIPS202_NATIVE_X86_64_KECCAK_F1600_X4_AVX2_H */

data/ext/pqcrypto/vendor/mldsa-native/mldsa/src/fips202/native/x86_64/src/fips202_native_x86_64.h

@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) The mlkem-native project authors
+ * Copyright (c) The mldsa-native project authors
+ * SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT
+ */
+
+#ifndef MLD_FIPS202_NATIVE_X86_64_SRC_FIPS202_NATIVE_X86_64_H
+#define MLD_FIPS202_NATIVE_X86_64_SRC_FIPS202_NATIVE_X86_64_H
+
+#include "../../../../cbmc.h"
+#include "../../../../common.h"
+
+/* TODO: Reconsider whether this check is needed -- x86_64 is always
+ * little-endian, so the backend selection already implies this. */
+#ifndef MLD_SYS_LITTLE_ENDIAN
+#error Expecting a little-endian platform
+#endif
+
+#define mld_keccakf1600_round_constants \
+  MLD_NAMESPACE(keccakf1600_round_constants)
+MLD_INTERNAL_DATA_DECLARATION const uint64_t
+    mld_keccakf1600_round_constants[24];
+
+#define mld_keccak_rho8 MLD_NAMESPACE(keccak_rho8)
+MLD_INTERNAL_DATA_DECLARATION const uint64_t mld_keccak_rho8[4];
+
+#define mld_keccak_rho56 MLD_NAMESPACE(keccak_rho56)
+MLD_INTERNAL_DATA_DECLARATION const uint64_t mld_keccak_rho56[4];
+
+#define mld_keccak_f1600_x4_avx2_asm MLD_NAMESPACE(keccak_f1600_x4_avx2_asm)
+void mld_keccak_f1600_x4_avx2_asm(uint64_t states[100], const uint64_t rc[24],
+                                  const uint64_t rho8[4],
+                                  const uint64_t rho56[4])
+/* This must be kept in sync with the HOL-Light specification
+ * in proofs/hol_light/x86_64/proofs/keccak_f1600_x4_avx2_asm.ml */
+__contract__(
+  requires(memory_no_alias(states, sizeof(uint64_t) * 25 * 4))
+  requires(rc == mld_keccakf1600_round_constants)
+  requires(rho8 == mld_keccak_rho8)
+  requires(rho56 == mld_keccak_rho56)
+  assigns(memory_slice(states, sizeof(uint64_t) * 25 * 4))
+);
+
+#endif /* !MLD_FIPS202_NATIVE_X86_64_SRC_FIPS202_NATIVE_X86_64_H */