arm neon qshl: Fix UQSHL to match hardware. Add extensive test vector…

…s. (#1256)

UQSHL was wrong in a variety of ways.  Most importantly, it did not
trim off the low byte for the shift as per the instruction spec:
"... shifts the element by a value from the least significant byte
of the corresponding element of the second source register..."
This was not detected in the tests, as the test vectors were limited
to small shift values that did not demonstrate the fault.

Behavior is fixed to be correct, per hardware.  Test vectors are
dramatically expanded, using both the previous small-range values as
well as additional full-range values.  Vectors generated on an ARMv9
system (Google Compute Engine C4A).

simde/arm/neon/qshl.h

@@ -187,22 +187,24 @@ simde_vqshlb_u8(uint8_t a, int8_t b) {
     #endif
   #else
     uint8_t r;
+    int8_t b8 = HEDLEY_STATIC_CAST(int8_t, b);
 
-    if (b < -7)
-      b = -7;
-
-    if (b <= 0) {
-      r = a >> -b;
-    } else if (b < 7) {
-      r = HEDLEY_STATIC_CAST(uint8_t, a << b);
-      if ((r >> b) != a) {
-        r = UINT8_MAX;
-      }
-    } else if (a == 0) {
+    if ((b8 <= -8) || (a == 0))
+    {
       r = 0;
-    } else {
+    }
+    else if (b8 >= 8)
+    {
       r = UINT8_MAX;
     }
+    else if (b8 <= 0) {
+      r = a >> -b8;
+    } else {
+      r = HEDLEY_STATIC_CAST(uint8_t, a << b8);
+      if ((r >> b8) != a) {
+        r = UINT8_MAX;
+      }
+    }
 
     return r;
   #endif
@@ -228,22 +230,24 @@ simde_vqshlh_u16(uint16_t a, int16_t b) {
     #endif
   #else
     uint16_t r;
+    int8_t b8 = HEDLEY_STATIC_CAST(int8_t, b);
 
-    if (b < -15)
-      b = -15;
-
-    if (b <= 0) {
-      r = a >> -b;
-    } else if (b < 15) {
-      r = HEDLEY_STATIC_CAST(uint16_t, a << b);
-      if ((r >> b) != a) {
-        r = UINT16_MAX;
-      }
-    } else if (a == 0) {
+    if ((b8 <= -16) || (a == 0))
+    {
       r = 0;
-    } else {
+    }
+    else if (b8 >= 16)
+    {
       r = UINT16_MAX;
     }
+    else if (b8 <= 0) {
+      r = a >> -b8;
+    } else {
+      r = HEDLEY_STATIC_CAST(uint16_t, a << b8);
+      if ((r >> b8) != a) {
+        r = UINT16_MAX;
+      }
+    }
 
     return r;
   #endif
@@ -269,22 +273,24 @@ simde_vqshls_u32(uint32_t a, int32_t b) {
     #endif
   #else
     uint32_t r;
+    int8_t b8 = HEDLEY_STATIC_CAST(int8_t, b);
 
-    if (b < -31)
-      b = -31;
-
-    if (b <= 0) {
-      r = HEDLEY_STATIC_CAST(uint32_t, a >> -b);
-    } else if (b < 31) {
-      r = a << b;
-      if ((r >> b) != a) {
-        r = UINT32_MAX;
-      }
-    } else if (a == 0) {
+    if ((b8 <= -32) || (a == 0))
+    {
       r = 0;
-    } else {
+    }
+    else if (b8 >= 32)
+    {
       r = UINT32_MAX;
     }
+    else if (b8 <= 0) {
+      r = a >> -b8;
+    } else {
+      r = HEDLEY_STATIC_CAST(uint32_t, a << b8);
+      if ((r >> b8) != a) {
+        r = UINT32_MAX;
+      }
+    }
 
     return r;
   #endif
@@ -310,22 +316,24 @@ simde_vqshld_u64(uint64_t a, int64_t b) {
     #endif
   #else
     uint64_t r;
+    int8_t b8 = HEDLEY_STATIC_CAST(int8_t, b);
 
-    if (b < -63)
-      b = -63;
-
-    if (b <= 0) {
-      r = a >> -b;
-    } else if (b < 63) {
-      r = HEDLEY_STATIC_CAST(uint64_t, a << b);
-      if ((r >> b) != a) {
-        r = UINT64_MAX;
-      }
-    } else if (a == 0) {
+    if ((b8 <= -64) || (a == 0))
+    {
       r = 0;
-    } else {
+    }
+    else if (b8 >= 64)
+    {
       r = UINT64_MAX;
     }
+    else if (b8 <= 0) {
+      r = a >> -b8;
+    } else {
+      r = HEDLEY_STATIC_CAST(uint64_t, a << b8);
+      if ((r >> b8) != a) {
+        r = UINT64_MAX;
+      }
+    }
 
     return r;
   #endif