arm64-expand code for nrv* EOF at de-compress is not quite uniform

WINDOWS_BACK requirement also interferes.
	modified:   ../misc/testsuite/upx_testsuite_1-expected_sha256sums.sh
	modified:   stub/src/arch/arm64/v8/nrv2d_d32.S
	modified:   stub/src/arch/arm64/v8/nrv2e_d32.S
	modified:   stub/src/arm64-expand.S
           plus generated *.h, *.map

src/stub/src/arch/arm64/v8/nrv2d_d32.S

@@ -82,7 +82,7 @@ ucl_nrv2d_decompress_32: .globl ucl_nrv2d_decompress_32  // ARM mode
    Actual decompressed length is stored through plen_dst.
    For SAFE mode: at call, *plen_dst must be allowed length of output buffer.
 */
-        PUSH3(x2,x3, lr)
+        PUSH3(x2,x3, lr)  // MATCH_93
 #define sp_DST0 0  /* stack offset of original dst */
         add srclim,src,len,uxtw  // srclim= eof_src;
 #if 1==SAFE  /*{*/
@@ -95,7 +95,7 @@ ucl_nrv2d_decompress_32: .globl ucl_nrv2d_decompress_32  // ARM mode
 
 #if defined(WINDOWS_BACK) || !defined(DAISY_CHAIN)  //{
 eof_n2d:
-        POP2(x3,x4)  // r3= orig_dst; r4= plen_dst
+        POP2(x3,x4)  // MATCH_93  r3= orig_dst; r4= plen_dst
         SUB2(src,srclim)  // 0 if actual src length equals expected length
         SUB2(dst,x3)  // actual dst length
         str dstw,[x4]
@@ -108,7 +108,7 @@ eof_n2d:
 #endif
 
         mov x0,x5  // result value
-        POP1(lr)
+        POP1(lr)  // MATCH_93
         ret
 #endif  // WINDOWS_BACK || ! DAISY_CHAIN }
 

src/stub/src/arch/arm64/v8/nrv2e_d32.S

@@ -83,8 +83,8 @@ ucl_nrv2e_decompress_32: .globl ucl_nrv2e_decompress_32  // ARM mode
    Actual decompressed length is stored through plen_dst.
    For SAFE mode: at call, *plen_dst must be allowed length of output buffer.
 */
-        PUSH1(lr)
-        PUSH2(x2,x3)
+        PUSH1(lr)  // MATCH_91
+        PUSH2(x2,x3)  // MATCH_92
 #define sp_DST0 0  /* stack offset of original dst */
         add srclim,src,len,uxtw  // srclim= eof_src;
 #if 1==SAFE  /*{*/
@@ -104,7 +104,7 @@ bad_src_n2e:  # return value will be 1
 
 #if defined(WINDOWS_BACK) || !defined(DAISY_CHAIN)  //{
 eof_n2e:
-        POP2(x3,x4)  // r3= orig_dst; r4= plen_dst
+        POP2(x3,x4)  // MATCH_92  r3= orig_dst; r4= plen_dst
         SUB2(src,srclim)  // 0 if actual src length equals expected length
         SUB2(dst,x3)  // actual dst length
         str dstw,[x4]
@@ -118,7 +118,7 @@ cache_n2e:  // unused label
 #endif
 
         mov x0,x5  // result value
-        POP1(lr)
+        POP1(lr)  // MATCH_91
         ret
 #endif  // WINDOS_BACK || ! DAISY_CHAIN }
 

src/stub/src/arm64-expand.S

@@ -97,10 +97,11 @@ decompress:  // (src *, cpr_len, dst *, &dstlen);
         src .req x0
         dst .req x2
         dstw .req w2
+
   .globl eof
-eof_nrv: // .globl eof_nrv .type eof_nrv,%function
-eof_n2e: // .globl eof_n2e .type eof_n2e,%function
-eof_n2d: // .globl eof_n2d .type eof_n2d,%function
+
+// sync_cache is done in tail of f_expand, after possible unfilter
+// NYI: eof_n2b, eof_n2d, eof_n2e should be unified.
 eof_n2b: // .globl eof_n2b .type eof_n2b,%function
 eof:  // MATCH_90  end of a compressed extent; need sync_cache after unfilter
         ldr x3,[sp],#NBPW  // &input_eof
@@ -112,6 +113,24 @@ eof:  // MATCH_90  end of a compressed extent; need sync_cache after unfilter
         ldr lr,[sp],#NBPW
         ret
 
+
+// WINDOWS_BACK compatibility seems to be broken
+// if POP3 replaces POP2+POP1 (MATCH_92, MATCH_91)
+srclim .req x7
+eof_n2d: // .globl eof_n2d
+eof_n2e: // .globl eof_n2e
+        POP2(x3,x4)  // MATCH_92  r3= orig_dst; r4= plen_dst
+        SUB2(src,srclim)  // 0 if actual src length equals expected length
+        SUB2(dst,x3)  // actual dst length
+        str dstw,[x4]
+        POP1(lr)  // MATCH_91
+        ret
+
+    .unreq src
+    .unreq dst
+    .unreq dstw
+    .unreq srclim
+
 eof_lzma: .globl elf_lzma
         mov sp,fp
         mov x5,x0  // save result value
@@ -124,9 +143,6 @@ eof_lzma: .globl elf_lzma
         mov x0,x5  // result value
         ret
 
-    .unreq src
-    .unreq dst
-
 #define M_NRV2B_LE32    2
 #define M_NRV2B_8    3
 #define M_NRV2D_LE32    5

src/stub/tmp/arm64-linux.elf-fold.map

@@ -57,8 +57,8 @@ EXP_HEAD        0x0000000000000000      0x104
  EXP_HEAD       0x0000000000000000      0x104 tmp/arm64-expand.o
                 0x0000000000000000                f_expand
 
-EXP_TAIL        0x0000000000000000       0xa8
- EXP_TAIL       0x0000000000000000       0xa8 tmp/arm64-expand.o
+EXP_TAIL        0x0000000000000000       0xc0
+ EXP_TAIL       0x0000000000000000       0xc0 tmp/arm64-expand.o
                 0x0000000000000000                eof
 
 NRV2E           0x0000000000000000       0xf4