[Bug tree-optimization/108692] New: [11/12/13 Regression] Miscompilation of orc_test.c since r11-5160

[Bug tree-optimization/108692] New: [11/12/13 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

            Bug ID: 108692
           Summary: [11/12/13 Regression] Miscompilation of orc_test.c
                    since r11-5160
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jakub at gcc dot gnu.org
  Target Milestone: ---

Since r11-5160-g9fc9573f9a5e9432e53c7d the following testcase distilled from
gstreamer orc's orc_test.c is miscompiled on aarch64-linux with -O2
-ftree-vectorize:
typedef struct _OrcExecutor {
  void *pad1;
  int n;
  int pad2, pad3, pad4;
  void *arrays[64];
  int params[64];
  int accumulators[4];
} OrcExecutor;
typedef union { int i; float f; signed short x2[2]; signed char x4[4]; }
orc_union32;

__attribute__((noipa)) void
_backup_orc_sad_nxm_u8 (OrcExecutor * restrict ex)
{
  int i;
  int j;
  int n = ex->n;
  int m = ex->params[12];
  const signed char * restrict ptr4;
  const signed char * restrict ptr5;
  orc_union32 var12 = { 0 };
  signed char var32;
  signed char var33;
  for (j = 0; j < m; j++) {
    ptr4 = ((void *)(((unsigned char *)(ex->arrays[4])) + (ex->params[4] *
j)));
    ptr5 = ((void *)(((unsigned char *)(ex->arrays[5])) + (ex->params[5] *
j)));
    for (i = 0; i < n; i++) {
      var32 = ptr4[i];
      var33 = ptr5[i];
      var12.i = var12.i + (((int)(unsigned char)var32 - (int)(unsigned
char)var33)<0 ? -((int)(unsigned char)var32 - (int)(unsigned char)var33) :
((int)(unsigned char)var32 - (int)(unsigned char)var33));
    }
  }
  ex->accumulators[0] = var12.i;
}

int
main ()
{
  if (__CHAR_BIT__ != 8 || __SIZEOF_INT__ != 4)
    return 0;
  OrcExecutor ex;
  int i, j;
  typedef signed char A[20 * 384 * 2 + 1];
  __attribute__((aligned (128))) A buf;
  signed char a[79 * 20] = {
    122, 23, -89, 127, 24, 52, 92, 72, 18, -84, 59, -54, 33, 118, 37, -108,
-93, -59, -92, -4, 21, -51, -57, -53, 81, 85, -64, -88, -120, 1, 62, 69, -1,
93, -117, 75, -118, -87, 101, 6, -55, 103, -98, -48, 36, -91, 114, 63, -121,
10, -84, 14, -35, -93, 119, -52, 78, 107, 5, 70, 57, -100, -31, -74, -16, -72,
21, -88, 54, 86, -5, -80, 121, -83, -28, -51, -48, -30, 106, -85, 119, 33, -1,
-68, -52, 34, -108, 51, -12, -101, 78, -14, -16, -43, 4, -1, 120, 67, -43, -1,
-54, 91, -47, 115, 53, 56, 104, -76, 100, -15, -109, -81, 107, -73, -60, 97,
-109, -79, -114, 17, -26, -105, 20, 81, 121, 16, 51, -78, 124, -52, 111, 74, 2,
97, -98, -25, -111, -26, 86, 16, 105, 82, 24, -65, -107, -110, 57, -31, -77,
108, -108, -119, -126, 22, -15, -8, 48, -107, 19, 66, 43, 99, 7, 127, 103, -13,
108, 8, 106, 20, 75, -28, -27, -25, 83, 86, 49, 81, 79, -94, 47, -86, 107, -96,
-32, 42, 38, 2, -61, 20, 22, 100, -14, -96, 95, 24, -5, -87, -15, -81, 127, 27,
-123, 72, -115, 26, 85, -104, -68, -13, -98, -67, 10, -61, -42, -62, 119, -23,
-83, -15, 97, 45, -58, -36, -98, -83, 50, 42, 91, -117, -79, 34, -73, 88, -89,
-7, 31, -78, 120, -61, -87, -7, 77, 43, 100, -68, -26, -88, 30, 43, 84, 83, 61,
-99, 95, 110, 111, 7, 85, 65, 86, 58, -107, -74, -16, -32, -37, -76, -56, 52,
73, 60, -34, -124, 57, 68, 87, 104, -60, 86, -27, -64, 18, -24, -12, 111, 96,
122, -102, -39, -46, 114, 59, 29, -2, -73, -1, 18, 52, 15, -11, 3, -113, 57, 4,
43, -67, -54, -106, -17, -115, -56, 2, -35, 61, 8, -121, 25, -21, -45, -110,
54, 110, -23, 34, -112, 85, 106, -110, 63, 32, 6, -34, -109, -5, 49, 38, -28,
-109, -43, -31, 22, -68, -71, -32, -18, -60, -81, 9, 24, -108, -128, -47, -12,
-116, -104, 118, 36, 28, 114, 2, -42, 35, -63, 0, -46, 8, 99, -5, 67, -103, 7,
-70, 69, -90, -104, 62, 50, 92, 112, 120, -84, 99, 62, -54, 106, -73, -77, 68,
-60, 71, -17, -101, -4, 79, -108, 37, -94, -33, -46, 58, 48, -127, -96, 5, 113,
-6, -128, -92, 77, 65, 23, -14, -42, -107, -85, 77, 36, 39, 94, 34, -16, -124,
40, 61, -13, -55, -106, -46, -4, -123, 68, 74, -44, 108, 27, -49, -46, -63, 75,
-106, 60, -127, 7, -17, 100, 80, 3, 14, 127, 37, -81, -105, 68, -76, 48, -21,
73, -97, 35, 53, 28, 91, 17, 51, 36, -29, -80, -59, 54, 32, -33, 55, -10, -40,
-100, -25, -49, -71, 89, 113, 44, 53, 68, -4, -51, -11, -123, -40, -16, 70,
125, 0, -99, -104, -111, -13, 76, -78, -71, 84, 83, -74, -57, 65, 22, -39, 120,
115, -19, -34, 60, 26, -45, 87, -59, 118, -44, -101, 81, 53, 53, -122, -72,
-112, 96, -13, 53, 36, -106, -6, -12, -112, 83, -22, 95, -44, -128, 5, 40, 100,
91, -2, 22, 15, 38, 51, -111, -109, -109, -86, -20, -4, 83, 126, -68, -95, 71,
-109, 49, -23, -32, 79, -10, -30, -16, -31, -62, 18, 109, 50, 121, 72, -20,
-80, 2, 67, -35, -38, -26, 37, -116, -116, -82, -106, -45, 8, -7, -73, -96,
-56, 95, -51, 42, 6, 9, -86, -14, 7, 13, -37, 124, 43, -118, -109, -54, -74,
-121, 90, -99, 92, -116, 117, 107, -114, -2, 20, 31, -121, 59, 62, -109, 21,
122, 125, -51, 40, 96, 73, -3, 28, 97, 5, -53, 32, 87, 124, -24, 112, -70, -93,
36, 95, 59, 115, -73, 12, -89, -118, 18, 66, -107, -90, -86, 80, 14, -32, -121,
-88, -7, -88, -117, 58, 73, -124, -125, -114, -27, -120, -17, -93, 90, -108,
-110, -117, 67, 80, -72, -61, 116, -24, 74, 84, -23, -120, -110, -86, 113,
-106, -49, 55, 26, -77, 120, -57, 91, 111, 80, -94, 4, -50, 47, 99, -26, 127,
-118, -101, 66, -42, -57, 37, -18, -125, 18, -17, 88, 88, 46, -98, 125, -59,
55, -42, 125, 9, -76, -81, -97, 29, 89, -122, -56, -128, -36, -84, -71, -65,
-103, -47, 11, -58, -27, 47, -20, -23, 50, 118, -105, -78, -25, 122, -107, 104,
-4, -97, -75, 100, 89, 0, -52, 19, -20, 98, 45, -54, -116, -37, -25, 77, -111,
66, -63, 29, 47, 85, -10, -123, -105, -83, -79, 99, -52, 98, 71, -73, 69, 126,
54, -19, 72, 14, -90, -108, 107, -37, -29, 107, -107, 109, 49, 126, -75, 14,
-43, -69, 42, -86, 83, -65, 26, 80, 48, -88, 29, 105, -127, -72, -84, -74, 60,
-81, 72, 11, 59, -27, 106, -51, 15, 41, 45, 45, -118, 92, -67, 33, 11, -51,
127, 39, -121, 84, -2, -71, 92, 45, -104, -113, -42, -102, -19, -105, 125, 55,
3, -75, 36, 27, 57, 58, -77, -77, -14, 38, -81, 95, -3, 30, -114, -55, 100, 52,
-92, -55, 5, -118, 39, -97, 117, 103, 51, -42, 38, 29, 34, 23, -83, -76, 116,
80, 64, -69, -77, 97, -15, -49, 7, -119, -89, -19, -104, -109, -31, 119, -90,
-64, 43, 7, 106, 115, 92, -9, -84, -97, -108, -83, 33, -15, -7, -89, -118, -67,
44, 74, -115, -87, -115, 106, 82, 27, 52, -104, -35, 99, -86, 47, 46, -88, 107,
29, 26, -108, 28, 114, -25, -65, -116, -86, -86, -14, 45, 68, 11, -42, 75, 98,
-69, -108, 105, 88, 59, -60, -32, -25, 120, -90, 65, 121, 42, -105, 33, -4,
120, 79, -88, -124, 106, -32, -40, -97, 18, 110, -98, 119, 122, -81, -101, -85,
76, 34, -81, -22, 68, 14, 73, 83, 86, 65, 113, -112, -98, -122, -110, -84, 12,
-29, 9, -72, -52, -105, 110, 34, 118, -42, -91, -48, -6, 66, -93, -1, 73, 41,
0, -33, -60, -127, 65, -22, -73, -111, 115, 47, 78, 39, 28, -21, -55, 53, -20,
-30, -22, 29, 11, 76, -105, -8, -52, -90, 84, 98, -123, 98, -124, 12, 49, 78,
95, -1, -58, -60, -78, -79, 87, -105, 90, -55, -22, -72, 126, -78, 23, -84,
-112, -31, 101, -82, 1, -25, 100, -93, -57, 54, -118, -62, -113, -77, -50, 0,
-62, -1, 6, 60, 58, -123, 1, 47, -63, 67, 100, 17, 77, 89, 69, 12, -2, 84,
-119, -77, 96, 48, 55, 109, 30, 85, 46, 22, -51, -17, -51, -105, -42, -113,
-109, 126, -38, 110, -58, 121, 126, 77, 31, -112, -6, 12, 52, 38, 97, 12, -87,
-82, -82, 70, 22, 44, -110, 40, -73, 114, 0, 51, -41, 24, 116, 86, -7, 56, 7,
8, -53, -18, 52, 39, -53, -109, 104, -109, -74, -76, 75, -119, -75, -44, -47,
-63, 71, -26, -1, 71, -19, -119, 125, 27, -105, 1, -31, -121, -26, -102, 125,
33, 51, 48, 13, -124, 122, 48, 97, -6, -53, 10, 113, 77, 93, -25, 95, -119, 5,
118, 83, -81, 116, 116, 23, 30, 51, -2, 113, -63, 96, -106, 68, -7, 38, -9,
101, 73, -3, -104, 62, -60, -89, 2, 37, 55, -26, -46, 122, 22, -3, -115, -125,
39, -19, 40, 7, -16, 108, 82, -84, -30, -91, -121, 16, 66, -77, -43, -86, -37,
-37, 109, 95, 101, -117, -23, -28, 80, 63, 89, -17, -96, 93, 68, -20, -12, 93,
66, 107, -122, -72, -77, 122, 50, 31, -97, -18, -9, 72, -103, 17, 46, 121, 80,
11, -120, -43, -11, 12, -55, 105, 81, 63, 70, -10, -94, 110, 31, 124, 57, 117,
61, -117, -44, 104, 10, 73, -68, -124, -13, 66, -98, -42, -97, -13, 42, -83,
-20, 20, 109, -63, -82, -87, -17, 32, 38, -57, -116, -36, 68, 39, 18, 118, -99,
108, -96, -15, 79, -61, -97, -76, -91, -6, 104, -122, 86, -110, 31, -56, -30,
83, 25, -1, -126, 47, -9, 81, -11, 30, 71, -65, 9, -71, -6, 35, -24, -104,
-118, -23, 40, 109, -24, -115, -112, -35, 38, -40, -85, 34, -23, -38, 15, 97,
-116, -90, 106, -52, 21, -48, 81, 119, -40, -89, -55, -5, -19, 51, -111, -51,
83, -19, 53, 92, -31, 65, -44, -20, -16, -60, -84, 53, -54, 53, 106, -70, -56,
-54, -67, 51, 111, 8, 47, -9, 47, 25, 81, 21, -112, -122, 82, 44, -62, 16, 13,
-94, 17, 125, -31, 80, 40, 35, 81, -19, 6, 54, 90, 59, -64, -47, -104, -13, -5,
-59, 23, 113, 40, 37, -6, 32, 126, -114, -50, -93, -71, 55, -85, -55, -108,
-68, 16, 46, 7, -86, 124, -80, 64, -55, -52, -104, 99, 45, 88, -123, 2, -5,
-52, 24, 91, -4, -58, 97, 84, -29, 52, -41, -80, 69, -87, -47, -122, -29, -103,
-18, -88, 127, 79, -32, -90, -79, -78, 125, -54, 58, -15, 57, -32, -28, 44,
103, -55, 23, -29, 47, -65, 59, 31, 78, 101, -47, 98, -111, -106, 81, 79, 15,
-22, -63, 26, 41, 8, 26, -82, 66, -82, 90, -97, 63, -127, 104, -42, -67, -107,
-84, -10, 21, -68, 106, -112, 17, -113, -29, 12, 107, 62, 42, 82, 9, 23, -57,
59, -5, -41, 13
  };
  signed char b[79 * 20] = {
    -13, -94, 73, 51, 103, 99, -63, -80, -90, -3, -30, 114, -88, 82, 59, -41,
-99, -21, 34, 92, -68, -39, 96, -36, -52, 77, -100, -122, 111, -9, -120, -97,
49, 31, -51, -3, 84, -124, -73, 63, 89, -9, -96, -51, 78, -73, 30, -109, -37,
-117, -20, 124, 10, -94, -103, -80, -39, 42, -14, -114, -126, -78, 48, 104,
-123, 64, -30, 1, 121, 50, -103, -57, -104, -43, 85, -47, 3, -57, -78, -13, 61,
105, -9, -97, 73, -83, -14, -113, 125, 6, 100, -63, 34, 75, 83, -16, -85, -101,
102, 89, -114, 6, 29, 75, -26, -4, -47, 113, -54, 118, 111, -32, 106, -26, -12,
-33, -17, -33, -88, 120, 17, -124, 26, -62, 117, -78, -33, -20, -52, -19, 9,
10, -84, 11, 105, -97, 44, -94, -123, -17, 66, 51, 88, 1, -35, 19, -52, 118,
-62, 29, -8, 13, 40, 25, 25, -15, 18, -21, 94, 103, 109, 98, -14, -82, -84, 74,
66, 108, 108, -36, 86, 81, -50, -57, 110, -71, -73, -41, -46, -127, 127, 53,
86, -83, 94, 100, -81, -68, -105, -10, 121, -59, 98, 18, 4, 90, -102, -88, -30,
25, 43, 110, 49, 17, 57, -95, -87, -37, 17, 125, 100, 117, -48, 120, 112, -35,
-43, 67, 23, 114, -24, 126, -23, 10, 39, 61, -114, -46, -76, 63, 77, -23, 32,
-24, 91, -69, -106, 52, 31, 83, 116, 96, -110, -67, -70, -63, -54, 22, -36,
101, -32, -60, 49, 24, 22, 117, -82, 37, -76, -53, 36, -20, -92, -60, 61, 40,
-51, -21, 53, -3, 47, 64, 88, -60, -61, -24, 62, -105, 16, -107, -55, -101,
-115, -77, -37, 16, -70, 88, -40, 26, -3, -24, -33, 76, 61, -117, -57, -6, 64,
80, -37, 59, -59, 30, 86, -64, 96, 95, -99, 88, 25, 106, 63, 63, -60, -76, 117,
103, 29, 45, -22, -11, -30, 65, 10, -105, 2, 99, -16, -51, 45, -109, -20, -62,
-41, -19, 83, 38, -22, -21, 13, -55, 45, 93, -36, -4, 114, 42, 108, -88, -126,
-60, 32, 43, -55, 108, -105, -10, -62, 43, 0, -25, 32, -116, 34, 103, -63, 26,
42, 67, 84, -99, 51, 76, -60, 19, -104, -83, 107, 112, -7, 15, -8, 21, 14, 110,
28, 18, 46, -71, -9, 28, -98, 44, 70, 33, 61, -55, -115, -114, -41, 26, -34,
31, -45, 50, 113, -10, 25, 8, -21, -21, -67, -37, -110, 92, -42, -36, -120,
-72, 30, -104, -59, -54, 75, 44, -42, -57, 48, 39, 87, 22, -110, 61, 49, -31,
-53, -79, -108, 96, 33, -128, -77, -32, -28, 78, 67, -106, 73, 18, -79, 37, 9,
-128, 38, -15, -71, 57, 59, 33, -24, -75, 124, 118, -6, -20, -98, -49, -57,
-57, -128, 120, -13, 74, 56, -93, 89, -108, 77, 56, 101, -109, -1, 126, -92, 6,
118, -59, 87, 59, 21, 8, -125, -97, 64, -58, -96, -4, -17, -94, 7, 115, -12,
-71, 25, 57, 12, -57, 47, 17, -89, -7, 68, -115, 119, -113, 100, 4, -69, 58,
-45, 82, 72, 47, -2, 83, 92, 14, 35, 29, 34, 118, -65, -1, 5, 100, -69, 61,
-75, 80, 85, -119, -6, 6, -49, 65, 59, 69, -38, 108, -51, 25, -63, -70, -104,
-53, 66, 39, 24, -62, 95, -87, -105, 90, 34, 114, -49, 25, -127, 106, -95, -50,
19, 0, -11, 56, 72, -79, 84, 121, 105, 89, 70, 57, 83, 60, 109, -103, -13, -45,
94, -34, -128, 88, 67, -38, 123, 16, 49, 26, 104, 92, 74, -47, 73, 125, -110,
-121, -33, 42, -110, -127, -96, -44, -81, -17, 119, -77, 71, -40, 62, -104,
-111, -58, -4, 114, 49, 49, -34, -99, -98, -84, -89, 27, 33, -105, 28, 119, 51,
65, 31, 32, 83, -60, -23, 56, 78, -65, -22, -119, -69, -19, -102, 98, -93, 106,
-3, 54, -19, -91, -7, -14, -116, 36, 88, 34, 103, -5, -101, -62, 11, 10, -124,
-27, 19, -7, 97, 8, 83, -11, 26, 51, -57, 39, -6, -43, 62, -7, 61, 93, -58, 8,
94, -104, -42, -25, 29, -51, 62, -103, 82, -102, -49, -51, 116, -115, -54, -41,
-24, -126, 9, -1, 46, 89, -109, 44, 85, -73, 59, 95, 24, -114, -18, -38, 1,
-10, -13, -107, -64, 76, 47, 15, 15, -31, 99, -43, 70, 101, 41, -45, -67, 18,
-84, 94, -27, -104, 42, -3, 37, -53, 43, -5, 118, 15, 126, -83, -67, 127, -8,
-125, -20, -32, -20, -47, -41, 72, 105, 91, -41, -114, 54, -93, 32, 112, -114,
-15, -30, -121, -100, 39, 35, -21, -75, 65, 87, -90, 88, -82, -60, -39, -62,
-49, 117, 54, 87, -78, 78, 76, 107, 127, 13, -97, 71, 40, 17, 31, 108, 71, 127,
-40, 95, -5, -14, 51, 56, -122, -32, -13, -23, -89, -54, -21, 79, -72, 43,
-128, -6, -120, 30, -29, -67, -10, -51, 10, -12, 82, -98, 115, -104, 115, 28,
-104, 4, -104, -91, 49, -124, -1, -39, 44, -93, 71, -66, 109, 49, -121, -77,
69, 22, 42, 26, -76, -38, -111, -59, 95, -16, 102, 29, -15, 0, 115, -128, -99,
-114, -8, 115, 66, 85, -50, -41, 46, -81, -98, 65, 109, 28, 86, -25, -118, 37,
-42, -37, 38, -12, 93, 5, 58, -80, -74, -31, 43, -102, 34, -2, -109, -23, 4,
50, -55, 101, 79, -98, 54, -58, -80, 101, 87, -54, 122, 43, -111, 24, 86, 79,
-72, -41, -89, -23, 94, 13, -66, -126, -37, -70, -62, -106, 33, -79, 93, -54,
61, 57, 39, -19, -84, 103, -45, -118, -64, 85, 113, -2, -23, -76, -78, -29,
-102, -45, -106, -111, -108, -97, 90, -103, 64, -48, 94, -111, -113, -115, 89,
-98, -110, -34, 93, -58, -40, -43, -113, -13, -117, 124, -94, -58, 73, 125, 24,
33, -60, -108, 56, 115, 24, -75, -50, -62, -17, -4, -91, 122, 87, 89, -59, -9,
94, -8, 108, -44, 115, 113, -49, -15, 124, -76, -115, 70, -74, -78, -9, 107,
80, -52, 3, -29, -57, -4, -114, 14, 25, 9, 122, -61, -124, 8, 80, 78, 26, -37,
111, -40, 19, 107, -16, -77, -20, 60, 114, 56, 23, -82, -94, 94, 98, -4, -116,
31, 39, -13, -91, -117, 57, 109, -108, -87, -119, -107, -60, 17, -100, -39,
-85, 74, -83, -83, -41, -28, -95, 6, -67, 85, -81, -111, 65, -108, -53, -73, 5,
119, -41, 82, 114, -51, -119, 44, -121, 59, 26, 60, 100, -41, -76, -97, -107,
-122, -103, -3, -20, -63, 72, 17, 28, 103, 116, -121, 113, -91, -88, -57, 124,
-6, 114, 114, 118, -84, -128, 46, 31, 116, -37, 49, 90, -24, -29, -114, -123,
-84, 14, 66, -102, 88, 96, -2, 7, 40, -87, -78, 77, 103, -127, -34, 90, -104,
4, -104, -89, 105, 78, 91, 17, 88, -75, -106, 121, 34, -55, -15, 26, -16, 12,
93, -83, 65, 113, -30, 62, -106, -105, 40, -52, -33, -62, -62, -92, -88, -41,
89, 75, 47, -61, -12, 93, -8, -118, 3, 114, -4, -101, 126, 84, 113, 31, -121,
15, 44, -59, 70, -92, -34, 75, -123, 114, 16, -53, 10, 100, -32, -53, -72, -37,
121, 2, 100, 90, 83, -21, 120, -12, 48, 12, 74, 118, -46, 3, 58, 84, 125, 85,
-115, 84, 47, -10, 12, -4, -94, -104, 92, -67, -2, -76, 28, -36, 94, 89, -63,
36, 9, -73, 49, 77, 109, 106, 32, 42, 28, 84, -71, -68, -112, 75, 101, -109,
88, 4, -125, -41, -22, -33, 33, 102, -8, -25, 20, 62, 55, -84, -90, -72, -17,
-34, -1, -123, 67, -82, 79, 105, 62, -27, -63, -71, -95, -64, 31, -107, 111,
-110, 69, -26, 52, 24, 71, 22, 25, -101, 98, -110, -87, -56, -93, 1, 97, 79,
89, 65, -6, 62, 60, 38, 117, 26, 90, -15, -22, -62, -59, -110, 17, 4, -2, -98,
-115, -4, 32, -52, -33, 9, -16, 120, 33, -127, 68, -16, 116, -42, 57, -77, -96,
96, 83, -5, -106, -67, -72, -124, 80, 124, 106, 31, 103, -114, 64, -11, 113,
109, -87, -100, 32, 10, 32, -36, 28, 119, 110, 62, -101, 89, -93, 88, -58, 11,
-46, -37, -52, -109, -113, 89, -62, 69, 42, -41, 65, 57, -88, 23, -66, 110,
-123, -94, -89, -69, 44, -105, 67, 121, 50, 54, 71, -14, 86, 43, 55, -120, -25,
-8, 61, -53, 123, 118, -80, 13, 84, 83, 109, 34, -8, -65, 19, 54, 41, -107, 8,
-103, -83, 52, 37, -14, -38, 30, 33, 67, -29, 83, -66, -96, -27, 107, -75, -81,
-79, 118, 98, -34, 87, -108, 48, 103, 119, 83, -62, 5, 71, -53, -26, 115, 99,
123, 59, -87, 88, -30, -100, -42, -17, -35, 83, 28, 57, -61, -1, 13, 95, 66,
65, -23, 38, -8, 41, -106, 69, -47, 99, 109, -95, 83, 77, -102, 74, -14, -19,
-126, -95, 71, -111, -73, 61, 57, 17, -126, -30, -9, -32, -80, 87, -95, 81, 31,
5, -73, -116, 12, 84, 66, 78, 26, -59, -60, 100, 33, -49, -32, -27, -68, 123,
-54, 50, -111, -66, 109, -81, 95, -102, 93, 62, -55, -13, -125, -128, -28, 50,
-126, -54, -126, -101, -21, -23, 60, 111, 38, -106, 65, -63, 27, -72, -31, 24,
-93, 92, -39, -54, -78
  };
  ex.n = 79;
  ex.params[4] = 384;
  ex.params[5] = 384;
  ex.params[12] = 20;
  ex.arrays[4] = &buf[0];
  ex.arrays[5] = &buf[20 * 384 + 1];
  for (i = 0; i < 20; ++i)
    {
      for (j = 0; j < 79; ++j)
        {
          buf[i * 384 + j] = a[i * 79 + j];
          buf[(i + 20) * 384 + j + 1] = b[i * 79 + j];
        }
    }
  _backup_orc_sad_nxm_u8 (&ex);
  if (ex.accumulators[0] != 136665)
    __builtin_abort ();
  return 0;
}

Instead of computing 136665 it computes 135989.
The assembly difference r11-5159 to r11-5160 is:
--- orc_test.s_ 2023-02-07 12:18:43.029183179 +0100
+++ orc_test.s  2023-02-07 12:19:07.704817488 +0100
@@ -44,9 +44,9 @@ _backup_orc_sad_nxm_u8:
        ldr     q2, [x6, x1]
        ldr     q3, [x5, x1]
        add     x1, x1, 16
-       uabdl2  v0.8h, v2.16b, v3.16b
-       uabal   v0.8h, v2.8b, v3.8b
-       uadalp  v1.4s, v0.8h
+       sabdl2  v0.8h, v2.16b, v3.16b
+       sabal   v0.8h, v2.8b, v3.8b
+       sadalp  v1.4s, v0.8h
        cmp     x1, x8
        bne     .L5
        addv    s1, v1.4s

[Bug tree-optimization/108692] [11/12/13 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
           Keywords|                            |wrong-code
   Last reconfirmed|                            |2023-02-07
                 CC|                            |ktkachov at gcc dot gnu.org,
                   |                            |rearnsha at gcc dot gnu.org,
                   |                            |rsandifo at gcc dot gnu.org
   Target Milestone|---                         |11.4
             Target|                            |aarch64-linux
           Priority|P3                          |P2
             Status|UNCONFIRMED                 |NEW

[Bug tree-optimization/108692] [11/12/13 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Before the r11-5160 change, the innermost loop was an unsigned char x 16 -> int
x 4
SAD_EXPR:
  <bb 7> [local count: 567644343]:
  # vect_var12_i_49.13_90 = PHI <vect_patt_46.22_103(7), { 0, 0, 0, 0 }(6)>
  # ivtmp.57_149 = PHI <ivtmp.57_150(7), 0(6)>
  vect_var32_32.16_95 = MEM <const vector(16) signed char> [(const signed char
*)vectp.15_91 + ivtmp.57_149 * 1];
  vect_var33_33.19_100 = MEM <const vector(16) signed char> [(const signed char
*)vectp.18_96 + ivtmp.57_149 * 1];
  vect_var32.20_101 = VIEW_CONVERT_EXPR<vector(16) unsigned
char>(vect_var32_32.16_95);
  vect_var33.21_102 = VIEW_CONVERT_EXPR<vector(16) unsigned
char>(vect_var33_33.19_100);
  vect_patt_46.22_103 = SAD_EXPR <vect_var32.20_101, vect_var33.21_102,
vect_var12_i_49.13_90>;
  ivtmp.57_150 = ivtmp.57_149 + 16;
  if (ivtmp.57_150 != _166)
    goto <bb 7>; [83.33%]
  else
    goto <bb 10>; [16.67%]
with vect_patt_46.22_103/vect_var12_i_49.13_90 type being vector(4) int.
After the change it is signed char x 16 -> int x 4 SAD_EXPR instead:
  <bb 7> [local count: 567644343]:
  # vect_var12_i_49.13_99 = PHI <vect_patt_39.20_110(7), { 0, 0, 0, 0 }(6)>
  # ivtmp.55_156 = PHI <ivtmp.55_157(7), 0(6)>
  vect_var32_32.16_104 = MEM <const vector(16) signed char> [(const signed char
*)vectp.15_100 + ivtmp.55_156 * 1];
  vect_var33_33.19_109 = MEM <const vector(16) signed char> [(const signed char
*)vectp.18_105 + ivtmp.55_156 * 1];
  vect_patt_39.20_110 = SAD_EXPR <vect_var32_32.16_104, vect_var33_33.19_109,
vect_var12_i_49.13_99>;
  ivtmp.55_157 = ivtmp.55_156 + 16;
  if (ivtmp.55_157 != _173)
    goto <bb 7>; [83.33%]
  else
    goto <bb 10>; [16.67%]

[Bug tree-optimization/108692] [11/12/13 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

[Bug tree-optimization/108692] [11/12/13 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, on simplified -O2 -ftree-vectorize testcase with trunk:
int
foo (signed char *x, signed char *y, int n)
{
  int i, r = 0;
  signed char a, b;
  for (i = 0; i < n; i++)
    {
      a = x[i];
      b = y[i];
      int c = (unsigned char) a - (unsigned char) b;
      r = r + (c < 0 ? -c : c);
    }
  return r;
}
everything looks ok to me until vect_recog_sad_pattern is called.
The interesting part of the loop in question is:
<bb 3> [local count: 955630225]:
# i_22 = PHI <i_20(6), 0(5)>
# r_23 = PHI <r_19(6), 0(5)>
...
a.0_5 = (unsigned char) a_15;
_6 = (int) a.0_5;
b.1_7 = (unsigned char) b_17;
_8 = (int) b.1_7;
c_18 = _6 - _8;
_9 = ABS_EXPR <c_18>;
r_19 = _9 + r_23;
with 15/17 having signed char type, 5/7 unsigned char and everything else int.
Now, when vect_recog_sad_pattern is called, it sees as diff_stmt_vinfo->stmt
patt_34 = (a.0_5) w- (b.1_7);
which is reasonable, abs_stmt_vinfo was
patt_30 = ABS_EXPR <patt_33>;
where 30 is signed short, patt_33 too set to (signed short) patt_34 and 34
unsigned short.  Still, the widening subtraction is done with zero extensions
from unsigned char operands to unsigned short.
But vect_recog_sad_pattern calls
1325      if (!vect_widened_op_tree (vinfo, diff_stmt_vinfo, MINUS_EXPR,
WIDEN_MINUS_EXPR,
1326                                 false, 2, unprom, &half_type))
1327        return NULL;
and instead of returning a.0_5 and b.1_7 as the unpromoted operands and
unsigned char as half_type, it returns a_15 and b_17 as the unpromoted operands
and signed char as half_type.
I'd think if in vect_widened_op_tree after the early checks rhs_code != code we
shouldn't look through further promotions and just accept what we have.
as

[Bug tree-optimization/108692] [11/12/13 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 54419
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54419&action=edit
gcc13-pr108692.patch

Untested fix.

[Bug tree-optimization/108692] [11/12/13 Regression] Miscompilation of orc_test.c since r11-5160

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:6ad1c1027628f094260037536f6b6fcdb63b5add

commit r13-5742-g6ad1c1027628f094260037536f6b6fcdb63b5add
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Wed Feb 8 18:41:21 2023 +0100

    vect-patterns: Fix up vect_widened_op_tree [PR108692]

    The following testcase is miscompiled on aarch64-linux since r11-5160.
    Given
      <bb 3> [local count: 955630225]:
      # i_22 = PHI <i_20(6), 0(5)>
      # r_23 = PHI <r_19(6), 0(5)>
    ...
      a.0_5 = (unsigned char) a_15;
      _6 = (int) a.0_5;
      b.1_7 = (unsigned char) b_17;
      _8 = (int) b.1_7;
      c_18 = _6 - _8;
      _9 = ABS_EXPR <c_18>;
      r_19 = _9 + r_23;
    ...
    where SSA_NAMEs 15/17 have signed char, 5/7 unsigned char and rest is int
    we first pattern recognize c_18 as
    patt_34 = (a.0_5) w- (b.1_7);
    which is still correct, 5/7 are unsigned char subtracted in wider type,
    but then vect_recog_sad_pattern turns it into
    SAD_EXPR <a_15, b_17, r_23>
    which is incorrect, because 15/17 are signed char and so it is
    sum of absolute signed differences rather than unsigned sum of
    absolute unsigned differences.
    The reason why this happens is that vect_recog_sad_pattern calls
    vect_widened_op_tree with MINUS_EXPR, WIDEN_MINUS_EXPR on the
    patt_34 = (a.0_5) w- (b.1_7); statement's vinfo and vect_widened_op_tree
    calls vect_look_through_possible_promotion on the operands of the
    WIDEN_MINUS_EXPR, which looks through the further casts.
    vect_look_through_possible_promotion has careful code to stop when there
    would be nested casts that need to be preserved, but the problem here
    is that the WIDEN_*_EXPR operation itself has an implicit cast on the
    operands already - in this case of WIDEN_MINUS_EXPR the unsigned char
    5/7 SSA_NAMEs are widened to unsigned short before the subtraction,
    and vect_look_through_possible_promotion obviously isn't told about that.

    Now, I think when we see those WIDEN_{MULT,MINUS,PLUS}_EXPR codes, we had
    to look through possible promotions already when creating those and so
    vect_look_through_possible_promotion again isn't really needed, all we need
    to do is arrange what that function will do if the operand isn't result
    of any cast.  Other option would be let
vect_look_through_possible_promotion
    know about the implicit promotion from the WIDEN_*_EXPR, but I'm afraid
    that would be much harder.

    2023-02-08  Jakub Jelinek  <jakub@redhat.com>

            PR tree-optimization/108692
            * tree-vect-patterns.cc (vect_widened_op_tree): If rhs_code is
            widened_code which is different from code, don't call
            vect_look_through_possible_promotion but instead just check op is
            SSA_NAME with integral type for which vect_is_simple_use is true
            and call set_op on this_unprom.

            * gcc.dg/pr108692.c: New test.

[Bug tree-optimization/108692] [11/12 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[11/12/13 Regression]       |[11/12 Regression]
                   |Miscompilation of           |Miscompilation of
                   |orc_test.c since r11-5160   |orc_test.c since r11-5160

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed on the trunk so far.

[Bug tree-optimization/108692] [11/12 Regression] Miscompilation of orc_test.c since r11-5160

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:00136f439e2849af2bfd9934d79a8297ab09a1d9

commit r12-9152-g00136f439e2849af2bfd9934d79a8297ab09a1d9
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Wed Feb 8 18:41:21 2023 +0100

    vect-patterns: Fix up vect_widened_op_tree [PR108692]

    The following testcase is miscompiled on aarch64-linux since r11-5160.
    Given
      <bb 3> [local count: 955630225]:
      # i_22 = PHI <i_20(6), 0(5)>
      # r_23 = PHI <r_19(6), 0(5)>
    ...
      a.0_5 = (unsigned char) a_15;
      _6 = (int) a.0_5;
      b.1_7 = (unsigned char) b_17;
      _8 = (int) b.1_7;
      c_18 = _6 - _8;
      _9 = ABS_EXPR <c_18>;
      r_19 = _9 + r_23;
    ...
    where SSA_NAMEs 15/17 have signed char, 5/7 unsigned char and rest is int
    we first pattern recognize c_18 as
    patt_34 = (a.0_5) w- (b.1_7);
    which is still correct, 5/7 are unsigned char subtracted in wider type,
    but then vect_recog_sad_pattern turns it into
    SAD_EXPR <a_15, b_17, r_23>
    which is incorrect, because 15/17 are signed char and so it is
    sum of absolute signed differences rather than unsigned sum of
    absolute unsigned differences.
    The reason why this happens is that vect_recog_sad_pattern calls
    vect_widened_op_tree with MINUS_EXPR, WIDEN_MINUS_EXPR on the
    patt_34 = (a.0_5) w- (b.1_7); statement's vinfo and vect_widened_op_tree
    calls vect_look_through_possible_promotion on the operands of the
    WIDEN_MINUS_EXPR, which looks through the further casts.
    vect_look_through_possible_promotion has careful code to stop when there
    would be nested casts that need to be preserved, but the problem here
    is that the WIDEN_*_EXPR operation itself has an implicit cast on the
    operands already - in this case of WIDEN_MINUS_EXPR the unsigned char
    5/7 SSA_NAMEs are widened to unsigned short before the subtraction,
    and vect_look_through_possible_promotion obviously isn't told about that.

    Now, I think when we see those WIDEN_{MULT,MINUS,PLUS}_EXPR codes, we had
    to look through possible promotions already when creating those and so
    vect_look_through_possible_promotion again isn't really needed, all we need
    to do is arrange what that function will do if the operand isn't result
    of any cast.  Other option would be let
vect_look_through_possible_promotion
    know about the implicit promotion from the WIDEN_*_EXPR, but I'm afraid
    that would be much harder.

    2023-02-08  Jakub Jelinek  <jakub@redhat.com>

            PR tree-optimization/108692
            * tree-vect-patterns.cc (vect_widened_op_tree): If rhs_code is
            widened_code which is different from code, don't call
            vect_look_through_possible_promotion but instead just check op is
            SSA_NAME with integral type for which vect_is_simple_use is true
            and call set_op on this_unprom.

            * gcc.dg/pr108692.c: New test.

    (cherry picked from commit 6ad1c1027628f094260037536f6b6fcdb63b5add)

[Bug tree-optimization/108692] [11 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[11/12 Regression]          |[11 Regression]
                   |Miscompilation of           |Miscompilation of
                   |orc_test.c since r11-5160   |orc_test.c since r11-5160

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed for gcc 12.3 too.

[Bug tree-optimization/108692] [11 Regression] Miscompilation of orc_test.c since r11-5160

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

--- Comment #8 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:eadc8fd8d7fd509b001fcdbe160e2d6862c745a4

commit r11-10709-geadc8fd8d7fd509b001fcdbe160e2d6862c745a4
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Wed Feb 8 18:41:21 2023 +0100

    vect-patterns: Fix up vect_widened_op_tree [PR108692]

    The following testcase is miscompiled on aarch64-linux since r11-5160.
    Given
      <bb 3> [local count: 955630225]:
      # i_22 = PHI <i_20(6), 0(5)>
      # r_23 = PHI <r_19(6), 0(5)>
    ...
      a.0_5 = (unsigned char) a_15;
      _6 = (int) a.0_5;
      b.1_7 = (unsigned char) b_17;
      _8 = (int) b.1_7;
      c_18 = _6 - _8;
      _9 = ABS_EXPR <c_18>;
      r_19 = _9 + r_23;
    ...
    where SSA_NAMEs 15/17 have signed char, 5/7 unsigned char and rest is int
    we first pattern recognize c_18 as
    patt_34 = (a.0_5) w- (b.1_7);
    which is still correct, 5/7 are unsigned char subtracted in wider type,
    but then vect_recog_sad_pattern turns it into
    SAD_EXPR <a_15, b_17, r_23>
    which is incorrect, because 15/17 are signed char and so it is
    sum of absolute signed differences rather than unsigned sum of
    absolute unsigned differences.
    The reason why this happens is that vect_recog_sad_pattern calls
    vect_widened_op_tree with MINUS_EXPR, WIDEN_MINUS_EXPR on the
    patt_34 = (a.0_5) w- (b.1_7); statement's vinfo and vect_widened_op_tree
    calls vect_look_through_possible_promotion on the operands of the
    WIDEN_MINUS_EXPR, which looks through the further casts.
    vect_look_through_possible_promotion has careful code to stop when there
    would be nested casts that need to be preserved, but the problem here
    is that the WIDEN_*_EXPR operation itself has an implicit cast on the
    operands already - in this case of WIDEN_MINUS_EXPR the unsigned char
    5/7 SSA_NAMEs are widened to unsigned short before the subtraction,
    and vect_look_through_possible_promotion obviously isn't told about that.

    Now, I think when we see those WIDEN_{MULT,MINUS,PLUS}_EXPR codes, we had
    to look through possible promotions already when creating those and so
    vect_look_through_possible_promotion again isn't really needed, all we need
    to do is arrange what that function will do if the operand isn't result
    of any cast.  Other option would be let
vect_look_through_possible_promotion
    know about the implicit promotion from the WIDEN_*_EXPR, but I'm afraid
    that would be much harder.

    2023-02-08  Jakub Jelinek  <jakub@redhat.com>

            PR tree-optimization/108692
            * tree-vect-patterns.c (vect_widened_op_tree): If rhs_code is
            widened_code which is different from code, don't call
            vect_look_through_possible_promotion but instead just check op is
            SSA_NAME with integral type for which vect_is_simple_use is true
            and call set_op on this_unprom.

            * gcc.dg/pr108692.c: New test.

    (cherry picked from commit 6ad1c1027628f094260037536f6b6fcdb63b5add)

[Bug tree-optimization/108692] [11 Regression] Miscompilation of orc_test.c since r11-5160

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108692

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed for 11.4 as well.