pq_crypto 0.4.2 → 0.5.0

data/ext/pqcrypto/vendor/mlkem-native/mlkem/src/native/aarch64/src/rej_uniform_asm.S

@@ -0,0 +1,226 @@
+/*
+ * Copyright (c) The mlkem-native project authors
+ * SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT
+ */
+
+/*yaml
+  Name: rej_uniform_asm
+  Description: Run rejection sampling on uniform random bytes to generate uniform random integers mod q
+  Signature: uint64_t mlk_rej_uniform_asm(int16_t r[256], const uint8_t *buf, unsigned buflen, const uint8_t table[2048])
+  ABI:
+    x0:
+      type: buffer
+      size_bytes: 512
+      permissions: write-only
+      c_parameter: int16_t r[256]
+      description: Output buffer
+    x1:
+      type: buffer
+      size_bytes: x2
+      permissions: read-only
+      c_parameter: const uint8_t *buf
+      description: Input buffer
+    x2:
+      type: scalar
+      c_parameter: unsigned buflen
+      description: Length of input buffer (must be multiple of 24)
+      test_with: 504  # MLKEM_GEN_MATRIX_NBLOCKS * MLK_XOF_RATE
+    x3:
+      type: buffer
+      size_bytes: 2048
+      permissions: read-only
+      c_parameter: const uint8_t table[2048]
+      description: Lookup table
+  Stack:
+    bytes: 576
+    description: register preservation and temporary storage
+*/
+
+#include "../../../common.h"
+#if defined(MLK_ARITH_BACKEND_AARCH64) && \
+    !defined(MLK_CONFIG_MULTILEVEL_NO_SHARED)
+
+/*
+ * WARNING: This file is auto-derived from the mlkem-native source file
+ *   dev/aarch64_opt/src/rej_uniform_asm.S using scripts/simpasm. Do not modify it directly.
+ */
+
+#if defined(__ELF__)
+.section .note.GNU-stack,"",@progbits
+#endif
+
+.text
+.balign 4
+.global MLK_ASM_NAMESPACE(rej_uniform_asm)
+MLK_ASM_FN_SYMBOL(rej_uniform_asm)
+
+        .cfi_startproc
+        sub sp, sp, #0x240
+        .cfi_adjust_cfa_offset 0x240
+        mov x7, #0x1                // =1
+        movk x7, #0x2, lsl #16
+        movk x7, #0x4, lsl #32
+        movk x7, #0x8, lsl #48
+        mov v31.d[0], x7
+        mov x7, #0x10               // =16
+        movk x7, #0x20, lsl #16
+        movk x7, #0x40, lsl #32
+        movk x7, #0x80, lsl #48
+        mov v31.d[1], x7
+        mov w11, #0xd01             // =3329
+        dup v30.8h, w11
+        mov x8, sp
+        mov x7, x8
+        mov x11, #0x0               // =0
+        eor v16.16b, v16.16b, v16.16b
+
+Lrej_uniform_initial_zero:
+        str q16, [x7], #0x40
+        stur q16, [x7, #-0x30]
+        stur q16, [x7, #-0x20]
+        stur q16, [x7, #-0x10]
+        add x11, x11, #0x20
+        cmp x11, #0x100
+        b.lt Lrej_uniform_initial_zero
+        mov x7, x8
+        mov x9, #0x0                // =0
+        mov x4, #0x100              // =256
+        cmp x2, #0x30
+        b.lo Lrej_uniform_loop48_end
+
+Lrej_uniform_loop48:
+        cmp x9, x4
+        b.hs Lrej_uniform_memory_copy
+        sub x2, x2, #0x30
+        ld3 { v0.16b, v1.16b, v2.16b }, [x1], #48
+        zip1 v4.16b, v0.16b, v1.16b
+        zip2 v5.16b, v0.16b, v1.16b
+        zip1 v6.16b, v1.16b, v2.16b
+        zip2 v7.16b, v1.16b, v2.16b
+        bic v4.8h, #0xf0, lsl #8
+        bic v5.8h, #0xf0, lsl #8
+        ushr v6.8h, v6.8h, #0x4
+        ushr v7.8h, v7.8h, #0x4
+        zip1 v16.8h, v4.8h, v6.8h
+        zip2 v17.8h, v4.8h, v6.8h
+        zip1 v18.8h, v5.8h, v7.8h
+        zip2 v19.8h, v5.8h, v7.8h
+        cmhi v4.8h, v30.8h, v16.8h
+        cmhi v5.8h, v30.8h, v17.8h
+        cmhi v6.8h, v30.8h, v18.8h
+        cmhi v7.8h, v30.8h, v19.8h
+        and v4.16b, v4.16b, v31.16b
+        and v5.16b, v5.16b, v31.16b
+        and v6.16b, v6.16b, v31.16b
+        and v7.16b, v7.16b, v31.16b
+        uaddlv s20, v4.8h
+        uaddlv s21, v5.8h
+        uaddlv s22, v6.8h
+        uaddlv s23, v7.8h
+        fmov w12, s20
+        fmov w13, s21
+        fmov w14, s22
+        fmov w15, s23
+        ldr q24, [x3, x12, lsl #4]
+        ldr q25, [x3, x13, lsl #4]
+        ldr q26, [x3, x14, lsl #4]
+        ldr q27, [x3, x15, lsl #4]
+        cnt v4.16b, v4.16b
+        cnt v5.16b, v5.16b
+        cnt v6.16b, v6.16b
+        cnt v7.16b, v7.16b
+        uaddlv s20, v4.8h
+        uaddlv s21, v5.8h
+        uaddlv s22, v6.8h
+        uaddlv s23, v7.8h
+        fmov w12, s20
+        fmov w13, s21
+        fmov w14, s22
+        fmov w15, s23
+        tbl v16.16b, { v16.16b }, v24.16b
+        tbl v17.16b, { v17.16b }, v25.16b
+        tbl v18.16b, { v18.16b }, v26.16b
+        tbl v19.16b, { v19.16b }, v27.16b
+        st1 { v16.8h }, [x7]
+        add x7, x7, x12, lsl #1
+        st1 { v17.8h }, [x7]
+        add x7, x7, x13, lsl #1
+        st1 { v18.8h }, [x7]
+        add x7, x7, x14, lsl #1
+        st1 { v19.8h }, [x7]
+        add x7, x7, x15, lsl #1
+        add x12, x12, x13
+        add x14, x14, x15
+        add x9, x9, x12
+        add x9, x9, x14
+        cmp x2, #0x30
+        b.hs Lrej_uniform_loop48
+
+Lrej_uniform_loop48_end:
+        cmp x9, x4
+        b.hs Lrej_uniform_memory_copy
+        cmp x2, #0x18
+        b.lo Lrej_uniform_memory_copy
+        sub x2, x2, #0x18
+        ld3 { v0.8b, v1.8b, v2.8b }, [x1], #24
+        zip1 v4.16b, v0.16b, v1.16b
+        zip1 v5.16b, v1.16b, v2.16b
+        bic v4.8h, #0xf0, lsl #8
+        ushr v5.8h, v5.8h, #0x4
+        zip1 v16.8h, v4.8h, v5.8h
+        zip2 v17.8h, v4.8h, v5.8h
+        cmhi v4.8h, v30.8h, v16.8h
+        cmhi v5.8h, v30.8h, v17.8h
+        and v4.16b, v4.16b, v31.16b
+        and v5.16b, v5.16b, v31.16b
+        uaddlv s20, v4.8h
+        uaddlv s21, v5.8h
+        fmov w12, s20
+        fmov w13, s21
+        ldr q24, [x3, x12, lsl #4]
+        ldr q25, [x3, x13, lsl #4]
+        cnt v4.16b, v4.16b
+        cnt v5.16b, v5.16b
+        uaddlv s20, v4.8h
+        uaddlv s21, v5.8h
+        fmov w12, s20
+        fmov w13, s21
+        tbl v16.16b, { v16.16b }, v24.16b
+        tbl v17.16b, { v17.16b }, v25.16b
+        st1 { v16.8h }, [x7]
+        add x7, x7, x12, lsl #1
+        st1 { v17.8h }, [x7]
+        add x7, x7, x13, lsl #1
+        add x9, x9, x12
+        add x9, x9, x13
+
+Lrej_uniform_memory_copy:
+        cmp x9, x4
+        csel x9, x9, x4, lo
+        mov x11, #0x0               // =0
+        mov x7, x8
+
+Lrej_uniform_final_copy:
+        ldr q16, [x7], #0x40
+        ldur q17, [x7, #-0x30]
+        ldur q18, [x7, #-0x20]
+        ldur q19, [x7, #-0x10]
+        str q16, [x0], #0x40
+        stur q17, [x0, #-0x30]
+        stur q18, [x0, #-0x20]
+        stur q19, [x0, #-0x10]
+        add x11, x11, #0x20
+        cmp x11, #0x100
+        b.lt Lrej_uniform_final_copy
+        mov x0, x9
+        b Lrej_uniform_return
+
+Lrej_uniform_return:
+        add sp, sp, #0x240
+        .cfi_adjust_cfa_offset -0x240
+        ret
+        .cfi_endproc
+
+MLK_ASM_FN_SIZE(rej_uniform_asm)
+
+#endif /* MLK_ARITH_BACKEND_AARCH64 && !MLK_CONFIG_MULTILEVEL_NO_SHARED */