[llvm] [NVVM] Add various intrinsic attrs, cleanup and consolidate td (PR #153436)

[Artem Belevich via llvm-commits]

================
@@ -1103,208 +1101,202 @@ let TargetPrefix = "nvvm" in {
     }
   }
 
-//
-// Add
-//
+  //
+  // Add
+  //
   let IntrProperties = [IntrNoMem, IntrSpeculatable, Commutative] in {
     foreach rnd = ["rn", "rz", "rm", "rp"] in {
       foreach ftz = ["", "_ftz"] in
         def int_nvvm_add_ # rnd # ftz # _f : NVVMBuiltin,
           DefaultAttrsIntrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty]>;
 
-    def int_nvvm_add_ # rnd # _d : NVVMBuiltin,
-        DefaultAttrsIntrinsic<[llvm_double_ty], [llvm_double_ty, llvm_double_ty]>;
+      def int_nvvm_add_ # rnd # _d : NVVMBuiltin,
+          DefaultAttrsIntrinsic<[llvm_double_ty], [llvm_double_ty, llvm_double_ty]>;
     }
   }
 
-//
-// Dot Product
-//
+  //
+  // Dot Product
+  //
   foreach a_type = ["s", "u"] in {
     foreach b_type = ["s", "u"] in {
       def int_nvvm_idp4a_ # a_type # _ # b_type :
-          DefaultAttrsIntrinsic<[llvm_i32_ty],
-              [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty],
-              [IntrNoMem, IntrSpeculatable]>;
+          PureIntrinsic<[llvm_i32_ty],
+              [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty]>;
       def int_nvvm_idp2a_ # a_type # _ # b_type :
-          DefaultAttrsIntrinsic<[llvm_i32_ty],
+          PureIntrinsic<[llvm_i32_ty],
             [llvm_i32_ty, llvm_i32_ty, llvm_i1_ty, llvm_i32_ty],
-            [IntrNoMem, IntrSpeculatable, ImmArg<ArgIndex<2>>]>;
+            [ImmArg<ArgIndex<2>>]>;
     }
   }
 
-//
-// Funnel-shift
-//
+  //
+  // Funnel-shift
+  //
   foreach direction = ["l", "r"] in
     def int_nvvm_fsh # direction # _clamp :
-      DefaultAttrsIntrinsic<[llvm_anyint_ty],
-        [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
-        [IntrNoMem, IntrSpeculatable]>;
+      PureIntrinsic<[llvm_anyint_ty],
+                    [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
 
-//
-// FLO - Find Leading One
-//
+  //
+  // FLO - Find Leading One
+  //
   foreach sign = ["s", "u"] in
     def int_nvvm_flo_ # sign :
-      DefaultAttrsIntrinsic<[llvm_i32_ty],
-        [llvm_anyint_ty, llvm_i1_ty],
-        [IntrNoMem, IntrSpeculatable, ImmArg<ArgIndex<1>>]>;
+      PureIntrinsic<[llvm_i32_ty], [llvm_anyint_ty, llvm_i1_ty],
+                    [ImmArg<ArgIndex<1>>]>;
 
-//
-// szext
-//
+  //
+  // szext
+  //
   foreach ext = ["sext", "zext"] in
     foreach mode = ["wrap", "clamp"] in
       def int_nvvm_ # ext # _ # mode :
-        DefaultAttrsIntrinsic<[llvm_i32_ty],
-          [llvm_i32_ty, llvm_i32_ty],
-          [IntrNoMem, IntrSpeculatable]>;
+        PureIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty]>;
 
-//
-// BMSK - bit mask
-//
+  //
+  // BMSK - bit mask
+  //
   foreach mode = ["wrap", "clamp"] in
     def int_nvvm_bmsk_ # mode :
-      DefaultAttrsIntrinsic<[llvm_i32_ty],
-        [llvm_i32_ty, llvm_i32_ty],
-        [IntrNoMem, IntrSpeculatable]>;
+      PureIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty]>;
 
-//
-// Convert
-//
-  let IntrProperties = [IntrNoMem, IntrSpeculatable] in {
-    def int_nvvm_lohi_i2d : NVVMBuiltin,
-        DefaultAttrsIntrinsic<[llvm_double_ty], [llvm_i32_ty, llvm_i32_ty]>;
+  //
+  // FNS - Find the n-th set bit
+  //
+  def int_nvvm_fns : NVVMBuiltin,
+      PureIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty]>;
 
-    def int_nvvm_d2i_lo : NVVMBuiltin,
-        DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_double_ty]>;
-    def int_nvvm_d2i_hi : NVVMBuiltin,
-        DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_double_ty]>;
+  //
+  // Convert
+  //
+  def int_nvvm_lohi_i2d : NVVMBuiltin,
+      PureIntrinsic<[llvm_double_ty], [llvm_i32_ty, llvm_i32_ty]>;
 
-    foreach rnd = ["rn", "rz", "rm", "rp"] in {
-      foreach ftz = ["", "_ftz"] in
-        def int_nvvm_d2f_ # rnd # ftz : NVVMBuiltin,
-            DefaultAttrsIntrinsic<[llvm_float_ty], [llvm_double_ty]>;
+  def int_nvvm_d2i_lo : NVVMBuiltin,
+      PureIntrinsic<[llvm_i32_ty], [llvm_double_ty]>;
+  def int_nvvm_d2i_hi : NVVMBuiltin,
+      PureIntrinsic<[llvm_i32_ty], [llvm_double_ty]>;
----------------
Artem-B wrote:

How sure are we that we want to allow speculative execution of FP conversions? Those may not always be fast. 
E.g. converting an integer that has more bits than mantissa can fit may potentially hit a slow path in SASS, if there's no complete support for the conversion in SASS.


https://github.com/llvm/llvm-project/pull/153436