Backends: Namespace assembly local labels with mld_ prefix

Previously, local labels in the assembly backends were prefixed only by
the name of the enclosing function, e.g. Lntt_layer123_start in
ntt_aarch64_asm.S. Global entry symbols are namespaced per project via
MLD_ASM_NAMESPACE(), but local labels are emitted as literal text and
survive verbatim into the autogenerated mldsa/*.S files.

mldsa-native derives its assembly from mlkem-native and kept the same
function and label names, so its local labels are byte-for-byte identical
(Lntt_layer123_start, Lrej_uniform_loop48, ...). This is harmless as long
as each backend is assembled into its own object file, but breaks when
both backends are merged into a single link unit: AWS-LC's FIPS build
concatenates all sources and runs them through the delocate tool, which
flattens them into one symbol namespace and panics on the duplicate
labels.

This commit adds a fixed mld_ project prefix to every local label across
all backends (AArch64 arithmetic and FIPS202, x86_64 AVX2, Armv8.1-M MVE),
not just the ones that clash today, so mldsa-native emits
Lmld_ntt_layer123_start and can no longer collide with the mlk_-prefixed
labels emitted by mlkem-native, independent of the integrator or platform.

The change is made in the dev/ assembly sources and propagated to
mldsa/*.S and the proofs/hol_light/* drops via scripts/simpasm. Local
labels emit no object-code bytes, and branch targets encode PC-relative
offsets rather than names, so the assembled .text is byte-identical; the
byte-locked HOL Light and CBMC proofs are unaffected.

check_asm_loop_labels in scripts/autogen is updated to require the
mld_-prefixed function-derived prefix, so a newly added un-namespaced
label fails CI.

Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>

Author: hanno-becker

dev/aarch64_clean/src/intt_aarch64_asm.S

@@ -272,7 +272,7 @@ MLD_ASM_FN_SYMBOL(intt_aarch64_asm)
         mov count, #16
 
         .p2align 2
-intt_layer5678_start:
+mld_intt_layer5678_start:
 
         ldr q_data0, [inp, #(16*0)]
         ldr q_data1, [inp, #(16*1)]
@@ -331,7 +331,7 @@ intt_layer5678_start:
         str q_data3, [inp, #(-64 + 16*3)]
 
         subs count, count, #1
-        cbnz count, intt_layer5678_start
+        cbnz count, mld_intt_layer5678_start
 
         // ---------------------------------------------------------------------
 
@@ -351,7 +351,7 @@ intt_layer5678_start:
 
         .p2align 2
 
-intt_layer1234_start:
+mld_intt_layer1234_start:
 
         ldr q_data0,  [in, #0]
         ldr q_data1,  [in, #(1*(512/8))]
@@ -464,7 +464,7 @@ intt_layer1234_start:
         str q_data7, [in, #(-16 + 7*(512/8))]
 
         subs count, count, #1
-        cbnz count, intt_layer1234_start
+        cbnz count, mld_intt_layer1234_start
 
         pop_stack
         ret

dev/aarch64_clean/src/mld_polyvecl_pointwise_acc_montgomery_l4_aarch64_asm.S

@@ -132,7 +132,7 @@ MLD_ASM_FN_SYMBOL(polyvecl_pointwise_acc_montgomery_l4_aarch64_asm)
 
     mov count, #(MLDSA_N / 4)
 
-polyvecl_pointwise_acc_montgomery_l4_loop_start:
+mld_polyvecl_pointwise_acc_montgomery_l4_loop_start:
     load_polys q_a_0, q_a_1, q_a_2, q_a_3, a_ptr, 0
     load_polys q_b_0, q_b_1, q_b_2, q_b_3, b_ptr, 0
     // Bounds: |a_{i}| < q, |b_{i}| < 9q
@@ -189,7 +189,7 @@ polyvecl_pointwise_acc_montgomery_l4_loop_start:
     str q_c_0, [out_ptr], #4*16
 
     subs count, count, #4
-    cbnz count, polyvecl_pointwise_acc_montgomery_l4_loop_start
+    cbnz count, mld_polyvecl_pointwise_acc_montgomery_l4_loop_start
 
     ret
 

dev/aarch64_clean/src/mld_polyvecl_pointwise_acc_montgomery_l5_aarch64_asm.S

@@ -132,7 +132,7 @@ MLD_ASM_FN_SYMBOL(polyvecl_pointwise_acc_montgomery_l5_aarch64_asm)
 
     mov count, #(MLDSA_N / 4)
 
-polyvecl_pointwise_acc_montgomery_l5_loop_start:
+mld_polyvecl_pointwise_acc_montgomery_l5_loop_start:
     load_polys q_a_0, q_a_1, q_a_2, q_a_3, a_ptr, 0
     load_polys q_b_0, q_b_1, q_b_2, q_b_3, b_ptr, 0
     // Bounds: |a_{i}| < q, |b_{i}| < 9q
@@ -199,7 +199,7 @@ polyvecl_pointwise_acc_montgomery_l5_loop_start:
     str q_c_0, [out_ptr], #4*16
 
     subs count, count, #4
-    cbnz count, polyvecl_pointwise_acc_montgomery_l5_loop_start
+    cbnz count, mld_polyvecl_pointwise_acc_montgomery_l5_loop_start
 
     ret
 

dev/aarch64_clean/src/mld_polyvecl_pointwise_acc_montgomery_l7_aarch64_asm.S

@@ -132,7 +132,7 @@ MLD_ASM_FN_SYMBOL(polyvecl_pointwise_acc_montgomery_l7_aarch64_asm)
 
     mov count, #(MLDSA_N / 4)
 
-polyvecl_pointwise_acc_montgomery_l7_loop_start:
+mld_polyvecl_pointwise_acc_montgomery_l7_loop_start:
     load_polys q_a_0, q_a_1, q_a_2, q_a_3, a_ptr, 0
     load_polys q_b_0, q_b_1, q_b_2, q_b_3, b_ptr, 0
     // Bounds: |a_{i}| < q, |b_{i}| < 9q
@@ -219,7 +219,7 @@ polyvecl_pointwise_acc_montgomery_l7_loop_start:
     str q_c_0, [out_ptr], #4*16
 
     subs count, count, #4
-    cbnz count, polyvecl_pointwise_acc_montgomery_l7_loop_start
+    cbnz count, mld_polyvecl_pointwise_acc_montgomery_l7_loop_start
 
     ret
 

dev/aarch64_clean/src/ntt_aarch64_asm.S

@@ -244,7 +244,7 @@ MLD_ASM_FN_SYMBOL(ntt_aarch64_asm)
 
         .p2align 2
 
-ntt_layer123_start:
+mld_ntt_layer123_start:
 
         ldr q_data0, [in, #0]
         ldr q_data1, [in, #(1*(1024/8))]
@@ -283,13 +283,13 @@ ntt_layer123_start:
         str q_data7, [in, #(-16 + 7*(1024/8))]
 
         subs count, count, #1
-        cbnz count, ntt_layer123_start
+        cbnz count, mld_ntt_layer123_start
 
         mov in, inp
         mov count, #8
 
         .p2align 2
-ntt_layer45678_start:
+mld_ntt_layer45678_start:
 
         ldr q_data0, [in, #(16*0)]
         ldr q_data1, [in, #(16*1)]
@@ -351,7 +351,7 @@ ntt_layer45678_start:
         str q_data7, [in, #(-16*1)]
 
         subs count, count, #1
-        cbnz count, ntt_layer45678_start
+        cbnz count, mld_ntt_layer45678_start
 
         pop_stack
         ret

dev/aarch64_clean/src/pointwise_montgomery_aarch64_asm.S

@@ -109,7 +109,7 @@ MLD_ASM_FN_SYMBOL(poly_pointwise_montgomery_aarch64_asm)
 
     mov count, #(MLDSA_N / 4)
 
-poly_pointwise_montgomery_loop_start:
+mld_poly_pointwise_montgomery_loop_start:
     ldr q_a_0, [a_ptr, #0*16]
     ldr q_a_1, [a_ptr, #1*16]
     ldr q_a_2, [a_ptr, #2*16]
@@ -143,7 +143,7 @@ poly_pointwise_montgomery_loop_start:
     str q_c_0, [a_ptr], #4*16
 
     subs count, count, #4
-    cbnz count, poly_pointwise_montgomery_loop_start
+    cbnz count, mld_poly_pointwise_montgomery_loop_start
 
     ret
 

dev/aarch64_clean/src/poly_caddq_aarch64_asm.S

@@ -47,7 +47,7 @@ MLD_ASM_FN_SYMBOL(poly_caddq_aarch64_asm)
         dup q_reg.4s, w9        // Load Q values
 
         mov count, #64/4
-poly_caddq_loop:
+mld_poly_caddq_loop:
         ldr q0, [a_ptr, #0*16]
         ldr q1, [a_ptr, #1*16]
         ldr q2, [a_ptr, #2*16]
@@ -64,7 +64,7 @@ poly_caddq_loop:
         str q0, [a_ptr], #4*16
 
         subs count, count, #1
-        bne poly_caddq_loop
+        bne mld_poly_caddq_loop
 
         ret
 

dev/aarch64_clean/src/poly_chknorm_aarch64_asm.S

@@ -55,7 +55,7 @@ MLD_ASM_FN_SYMBOL(poly_chknorm_aarch64_asm)
 
         mov count, #(64/4)
 
-poly_chknorm_loop:
+mld_poly_chknorm_loop:
         ldr q1, [a_ptr, #1*16]
         ldr q2, [a_ptr, #2*16]
         ldr q3, [a_ptr, #3*16]
@@ -67,7 +67,7 @@ poly_chknorm_loop:
         chknorm v0
 
         subs count, count, #1
-        bne poly_chknorm_loop
+        bne mld_poly_chknorm_loop
 
         // Return 1 if any of the 4 lanes is 0xffffffff
         umaxv s21, flags.4s

dev/aarch64_clean/src/poly_decompose_32_aarch64_asm.S

@@ -126,7 +126,7 @@ MLD_ASM_FN_SYMBOL(poly_decompose_32_aarch64_asm)
 
         mov count, #(64/4)
 
-poly_decompose_32_loop:
+mld_poly_decompose_32_loop:
         ldr q0, [a0_ptr, #0*16]
         ldr q1, [a0_ptr, #1*16]
         ldr q2, [a0_ptr, #2*16]
@@ -147,7 +147,7 @@ poly_decompose_32_loop:
         str q0, [a0_ptr], #4*16
 
         subs count, count, #1
-        bne poly_decompose_32_loop
+        bne mld_poly_decompose_32_loop
 
         ret
 

dev/aarch64_clean/src/poly_decompose_88_aarch64_asm.S

@@ -124,7 +124,7 @@ MLD_ASM_FN_SYMBOL(poly_decompose_88_aarch64_asm)
         dup barrett_const.4s, w11
 
         mov count, #(64/4)
-poly_decompose_88_loop:
+mld_poly_decompose_88_loop:
         ldr q0, [a0_ptr, #0*16]
         ldr q1, [a0_ptr, #1*16]
         ldr q2, [a0_ptr, #2*16]
@@ -145,7 +145,7 @@ poly_decompose_88_loop:
         str q0, [a0_ptr], #4*16
 
         subs count, count, #1
-        bne poly_decompose_88_loop
+        bne mld_poly_decompose_88_loop
 
         ret