Fix pixel_ssim_end4 asm function for x86_64 systems

diff --git a/common/x86/pixel-a.asm b/common/x86/pixel-a.asm
index 85e7d5a45c6303f7353cc77a89b5fe0e137b36d7..edadad38689abc1713fc61ab52829143b993e002 100644 (file)
--- a/common/x86/pixel-a.asm
+++ b/common/x86/pixel-a.asm
@@ -4689,7 +4689,8 @@ cglobal pixel_ssim_4x4x2_core, 4,4,8
 ;-----------------------------------------------------------------------------
 ; float pixel_ssim_end( int sum0[5][4], int sum1[5][4], int width )
 ;-----------------------------------------------------------------------------
-cglobal pixel_ssim_end4, 3,3,7
+cglobal pixel_ssim_end4, 2,3,7
+    mov      r2d, r2m
     movdqa    m0, [r0+ 0]
     movdqa    m1, [r0+16]
     movdqa    m2, [r0+32]

diff --git a/tools/checkasm.c b/tools/checkasm.c
index f72b7a00e5c47056d6814bafe732bf71e5ced065..cb889660c6e9d90f758a8971118cf7c213870d33 100644 (file)
--- a/tools/checkasm.c
+++ b/tools/checkasm.c
@@ -728,11 +728,14 @@ static int check_pixel( int cpu_ref, int cpu_new )
             fprintf( stderr, "ssim: %.7f != %.7f [FAILED]\n", res_c, res_a );
         }
         set_func_name( "ssim_core" );
-        call_c2( pixel_c.ssim_4x4x2_core,   pbuf1+2, (intptr_t)32, pbuf2+2, (intptr_t)32, sums );
-        call_a2( pixel_asm.ssim_4x4x2_core, pbuf1+2, (intptr_t)32, pbuf2+2, (intptr_t)32, sums );
+        call_c( pixel_c.ssim_4x4x2_core,   pbuf1+2, (intptr_t)32, pbuf2+2, (intptr_t)32, sums );
+        call_a( pixel_asm.ssim_4x4x2_core, pbuf1+2, (intptr_t)32, pbuf2+2, (intptr_t)32, sums );
         set_func_name( "ssim_end" );
         call_c2( pixel_c.ssim_end4,   sums, sums, 4 );
         call_a2( pixel_asm.ssim_end4, sums, sums, 4 );
+        /* check incorrect assumptions that 32-bit ints are zero-extended to 64-bit */
+        call_c1( pixel_c.ssim_end4,   sums, sums, 3 );
+        call_a1( pixel_asm.ssim_end4, sums, sums, 3 );
         report( "ssim :" );
     }