From f2a1dddbca2c0c6274b89dd708aefec5444c852b Mon Sep 17 00:00:00 2001
From: mykola2312 <49044616+mykola2312@users.noreply.github.com>
Date: Wed, 11 May 2022 19:18:45 +0300
Subject: [PATCH] implement cu_memtest in x86-64 assembly in more efficient way
 that C does

---
 arch/x86_64.S | 29 +++++++++++++++++++++++++++++
 cutil.c       |  4 ++--
 cutil.h       |  2 +-
 3 files changed, 32 insertions(+), 3 deletions(-)

diff --git a/arch/x86_64.S b/arch/x86_64.S
index 2bc995f..207caa0 100644
--- a/arch/x86_64.S
+++ b/arch/x86_64.S
@@ -1,6 +1,7 @@
 .globl hw_bswap16
 .globl hw_bswap32
 .globl hw_bswap64
+.globl cu_memtest
 
 .text
 
@@ -19,3 +20,31 @@ hw_bswap64:
     bswap %rax
     ret
 
+cu_memtest:
+    // RDI - ptr, RSI - size
+    xchg %rsi, %rdi
+    
+    mov %rdi, %rcx
+    shr $3, %rcx
+    je .testok
+.testq:
+    lodsq
+    test %rax, %rax
+    jnz .testfail
+    loop .testq
+
+    mov %rdi, %rcx
+    and $7, %rcx
+    je .testok
+.testb:
+    lodsb
+    test %al, %al
+    jnz .testfail
+    loop .testb
+
+.testok:
+    xor %rax, %rax
+    ret
+.testfail:
+    mov %rsi, %rax
+    ret
\ No newline at end of file
diff --git a/cutil.c b/cutil.c
index 4c8c170..ef5ca3d 100644
--- a/cutil.c
+++ b/cutil.c
@@ -25,7 +25,7 @@ void cutil_exit()
 {
 }
 
-const void* cu_memtest(const void* mem, uint size)
+/*const void* cu_memtest(const void* mem, uint size)
 {
     unsigned i;
     for (i = 0; i < size / sizeof(uword); i++)
@@ -41,4 +41,4 @@ const void* cu_memtest(const void* mem, uint size)
     }
 
     return NULL;
-}
+}*/
diff --git a/cutil.h b/cutil.h
index 14de1cb..f6e2b43 100644
--- a/cutil.h
+++ b/cutil.h
@@ -21,6 +21,6 @@ extern void* (*cu_memmove)(void*,const void*,size_t);
 void cutil_init();
 void cutil_exit();
 
-const void* cu_memtest(const void* mem, uint size);
+extern const void* cu_memtest(const void* mem, uint size);
 
 #endif
\ No newline at end of file