1 files changed, 127 insertions, 80 deletions

diff --git a/newlib/libc/machine/riscv/strcmp.S b/newlib/libc/machine/riscv/strcmp.S
index 12c39db..0b1dfc4 100644
--- a/newlib/libc/machine/riscv/strcmp.S
+++ b/newlib/libc/machine/riscv/strcmp.S
@@ -16,15 +16,15 @@
 .type  strcmp, @function
 strcmp:
 #if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)
-1:
+.Lcompare:
   lbu   a2, 0(a0)
   lbu   a3, 0(a1)
-  add   a0, a0, 1
-  add   a1, a1, 1
-  bne   a2, a3, 2f
-  bnez  a2, 1b
+  addi   a0, a0, 1
+  addi   a1, a1, 1
+  bne   a2, a3, .Lreturn_diff
+  bnez  a2, .Lcompare
 
-2:
+.Lreturn_diff:
   sub   a0, a2, a3
   ret
 
@@ -48,12 +48,16 @@ strcmp:
     REG_L a2, \i*SZREG(a0)
     REG_L a3, \i*SZREG(a1)
 
-    and   t0, a2, a5
-    or    t1, a2, a5
-    add   t0, t0, a5
-    or    t0, t0, t1
+    #if __riscv_zbb
+      orc.b a4, a2
+    #else
+      and   a4, a2, a5
+      or    t1, a2, a5
+      add   a4, a4, a5
+      or    a4, a4, t1
+    #endif
 
-    bne   t0, t2, .Lnull\i
+    bne   a4, t2, .Lnull\i
     .if \i+1-\n
       bne   a2, a3, .Lmismatch
     .else
@@ -95,73 +99,109 @@ strcmp:
 
 .Lmismatch:
   # words don't match, but a2 has no null byte.
+  #if __riscv_zbb
+    xor a4, a2, a3              # find differing bits
+
+    # Check system endianness
+    # If little-endian, use Count Trailing Zeros (ctz)
+    # If big-endian, use Count Leading Zeros (clz)
+    # This helps identify the position of the first differing byte between a2 and a3.
+
+    # For example, in little-endian, least significant byte comes first.
+    # So trailing zeros help find which byte position differs.
+
+    # In big-endian, most significant byte comes first, so leading zeros are used.
+    # The position will then be used to extract the differing byte.
+
+    #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+      ctz a5, a4
+    #else
+      clz a5, a4
+    #endif
+
+    andi a5, a5, -8             # find position of bit offset to the start of the byte where the first difference occurs
+
+
+    # Shift a2 and a3 right by a5 bits to bring the target byte to the LSB, and isolate the byte of interest
+    srl a2, a2, a5
+    and a2, a2, 0xff
+
+    srl a3, a3, a5
+    and a3, a3, 0xff
+
+
+    sub a0, a2, a3              # Calculate and return the difference in the isolated bytes
+    ret
+
+  #else
+    #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+      #if __riscv_xlen == 64
+        sll   a4, a2, 48
+        sll   a5, a3, 48
+        bne   a4, a5, .Lmismatch_upper
+        sll   a4, a2, 32
+        sll   a5, a3, 32
+        bne   a4, a5, .Lmismatch_upper
+      #endif
+      sll   a4, a2, 16
+      sll   a5, a3, 16
+      bne   a4, a5, .Lmismatch_upper
+
+      srl   a4, a2, 8*SZREG-16
+      srl   a5, a3, 8*SZREG-16
+      sub   a0, a4, a5
+      and   a1, a0, 0xff
+      bnez  a1, .Lfinal_upper_diff
+      ret
 
-#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
-
-#if __riscv_xlen == 64
-  sll   a4, a2, 48
-  sll   a5, a3, 48
-  bne   a4, a5, .Lmismatch_upper
-  sll   a4, a2, 32
-  sll   a5, a3, 32
-  bne   a4, a5, .Lmismatch_upper
-#endif
-  sll   a4, a2, 16
-  sll   a5, a3, 16
-  bne   a4, a5, .Lmismatch_upper
-
-  srl   a4, a2, 8*SZREG-16
-  srl   a5, a3, 8*SZREG-16
-  sub   a0, a4, a5
-  and   a1, a0, 0xff
-  bnez  a1, 1f
-  ret
-
-.Lmismatch_upper:
-  srl   a4, a4, 8*SZREG-16
-  srl   a5, a5, 8*SZREG-16
-  sub   a0, a4, a5
-  and   a1, a0, 0xff
-  bnez  a1, 1f
-  ret
-
-1:and   a4, a4, 0xff
-  and   a5, a5, 0xff
-  sub   a0, a4, a5
-  ret
-
-#else
-
-#if __riscv_xlen == 64
-  srl   a4, a2, 48
-  srl   a5, a3, 48
-  bne   a4, a5, .Lmismatch_lower
-  srl   a4, a2, 32
-  srl   a5, a3, 32
-  bne   a4, a5, .Lmismatch_lower
-#endif
-  srl   a4, a2, 16
-  srl   a5, a3, 16
-  bne   a4, a5, .Lmismatch_lower
-
-  srl	a4, a2, 8
-  srl   a5, a3, 8
-  bne   a4, a5, 1f
-  and   a4, a2, 0xff
-  and   a5, a3, 0xff
-1:sub	a0, a4, a5
-  ret
-
-.Lmismatch_lower:
-  srl	a2, a4, 8
-  srl   a3, a5, 8
-  bne   a2, a3, 1f
-  and   a2, a4, 0xff
-  and   a3, a5, 0xff
-1:sub	a0, a2, a3
-  ret
-
-#endif
+      .Lmismatch_upper:
+        srl   a4, a4, 8*SZREG-16
+        srl   a5, a5, 8*SZREG-16
+        sub   a0, a4, a5
+        and   a1, a0, 0xff
+        bnez  a1, .Lfinal_upper_diff
+        ret
+
+      .Lfinal_upper_diff:
+        and   a4, a4, 0xff
+        and   a5, a5, 0xff
+        sub   a0, a4, a5
+        ret
+    #else
+      #if __riscv_xlen == 64
+        srl   a4, a2, 48
+        srl   a5, a3, 48
+        bne   a4, a5, .Lmismatch_lower
+        srl   a4, a2, 32
+        srl   a5, a3, 32
+        bne   a4, a5, .Lmismatch_lower
+      #endif
+      srl   a4, a2, 16
+      srl   a5, a3, 16
+      bne   a4, a5, .Lmismatch_lower
+
+      srl	  a4, a2, 8
+      srl   a5, a3, 8
+      bne   a4, a5, .Lbyte_diff
+      and   a4, a2, 0xff
+      and   a5, a3, 0xff
+
+      .Lbyte_diff:
+        sub	a0, a4, a5
+        ret
+
+      .Lmismatch_lower:
+        srl	  a2, a4, 8
+        srl   a3, a5, 8
+        bne   a2, a3, .Lfinal_lower_diff
+        and   a2, a4, 0xff
+        and   a3, a5, 0xff
+
+      .Lfinal_lower_diff:
+        sub	a0, a2, a3
+        ret
+    #endif
+  #endif
 	
 .Lmisaligned:
   # misaligned
@@ -169,10 +209,10 @@ strcmp:
   lbu   a3, 0(a1)
   add   a0, a0, 1
   add   a1, a1, 1
-  bne   a2, a3, 1f
+  bne   a2, a3, .Lmisaligned_diff
   bnez  a2, .Lmisaligned
 
-1:
+.Lmisaligned_diff:
   sub   a0, a2, a3
   ret
 
@@ -188,9 +228,16 @@ strcmp:
   foundnull 1 3
   foundnull 2 3
 #endif
+#ifdef __riscv_cmodel_large
+  # Put the data within the funciton for large code model to prevent
+  # the data put too far.
+.align 3
+mask:
+.dword 0x7f7f7f7f7f7f7f7f
+#endif
 .size	strcmp, .-strcmp
 
-#if SZREG == 8
+#if SZREG == 8 && !defined(__riscv_cmodel_large)
 .section .srodata.cst8,"aM",@progbits,8
 .align 3
 mask: