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

[Alex MacLean via llvm-commits]

================
@@ -903,90 +914,79 @@ let TargetPrefix = "nvvm" in {
     }
   }
 
-//
-// Sad
-//
-  let IntrProperties = [IntrNoMem, IntrSpeculatable] in {
-    foreach sign = ["", "u"] in {
-      def int_nvvm_sad_ # sign # s : NVVMBuiltin,
-          DefaultAttrsIntrinsic<[llvm_i16_ty], [llvm_i16_ty, llvm_i16_ty, llvm_i16_ty]>;
+  //
+  // Sad - Sum of Absolute Differences
+  //
+  foreach sign = ["", "u"] in {
+    def int_nvvm_sad_ # sign # s : NVVMBuiltin,
+        PureIntrinsic<[llvm_i16_ty], [llvm_i16_ty, llvm_i16_ty, llvm_i16_ty]>;
 
-      def int_nvvm_sad_ # sign # i : NVVMBuiltin,
-          DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty]>;
+    def int_nvvm_sad_ # sign # i : NVVMBuiltin,
+        PureIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty]>;
 
-      def int_nvvm_sad_ # sign # ll : NVVMBuiltin,
-          DefaultAttrsIntrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_i64_ty, llvm_i64_ty]>;
-    }
+    def int_nvvm_sad_ # sign # ll : NVVMBuiltin,
+        PureIntrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_i64_ty, llvm_i64_ty]>;
   }
 
-//
-// Floor  Ceil
-//
-  let IntrProperties = [IntrNoMem, IntrSpeculatable] in {
-    foreach op = ["floor", "ceil"] in {
-      foreach ftz = ["", "_ftz"] in
-        def int_nvvm_ # op # ftz # _f : NVVMBuiltin,
-            DefaultAttrsIntrinsic<[llvm_float_ty], [llvm_float_ty]>;
-      def int_nvvm_ # op # _d : NVVMBuiltin,
-          DefaultAttrsIntrinsic<[llvm_double_ty], [llvm_double_ty]>;
-    }
+  //
+  // Floor  Ceil
+  //
+  foreach op = ["floor", "ceil"] in {
+    foreach ftz = ["", "_ftz"] in
+      def int_nvvm_ # op # ftz # _f : NVVMBuiltin,
+          PureIntrinsic<[llvm_float_ty], [llvm_float_ty]>;
+    def int_nvvm_ # op # _d : NVVMBuiltin,
+        PureIntrinsic<[llvm_double_ty], [llvm_double_ty]>;
   }
 
-//
-// Abs
-//
+  //
+  // Abs
+  //
   foreach ftz = ["", "_ftz"] in
     def int_nvvm_fabs # ftz :
-      DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>],
-                            [IntrNoMem, IntrSpeculatable]>;
+      PureIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
 
-//
-// Abs, Neg bf16, bf16x2
-//
+  //
+  // Neg bf16, bf16x2
+  //
   def int_nvvm_neg_bf16 : NVVMBuiltin,
-    DefaultAttrsIntrinsic<[llvm_bfloat_ty], [llvm_bfloat_ty], [IntrNoMem]>;
+    PureIntrinsic<[llvm_bfloat_ty], [llvm_bfloat_ty]>;
   def int_nvvm_neg_bf16x2 : NVVMBuiltin,
-    DefaultAttrsIntrinsic<[llvm_v2bf16_ty], [llvm_v2bf16_ty], [IntrNoMem]>;
+    PureIntrinsic<[llvm_v2bf16_ty], [llvm_v2bf16_ty]>;
 
-//
-// Round
-//
-  let IntrProperties = [IntrNoMem, IntrSpeculatable] in {
-    foreach ftz = ["", "_ftz"] in
-      def int_nvvm_round # ftz # _f : NVVMBuiltin,
-          DefaultAttrsIntrinsic<[llvm_float_ty], [llvm_float_ty]>;
+  //
+  // Round
+  //
+  foreach ftz = ["", "_ftz"] in
+    def int_nvvm_round # ftz # _f : NVVMBuiltin,
+        PureIntrinsic<[llvm_float_ty], [llvm_float_ty]>;
 
-    def int_nvvm_round_d : NVVMBuiltin,
-        DefaultAttrsIntrinsic<[llvm_double_ty], [llvm_double_ty]>;
-  }
+  def int_nvvm_round_d : NVVMBuiltin,
+      PureIntrinsic<[llvm_double_ty], [llvm_double_ty]>;
 
-//
-// Trunc
-//
-  let IntrProperties = [IntrNoMem, IntrSpeculatable] in {
-    foreach ftz = ["", "_ftz"] in
-      def int_nvvm_trunc # ftz # _f : NVVMBuiltin,
-          DefaultAttrsIntrinsic<[llvm_float_ty], [llvm_float_ty]>;
+  //
+  // Trunc
+  //
+  foreach ftz = ["", "_ftz"] in
+    def int_nvvm_trunc # ftz # _f : NVVMBuiltin,
+        PureIntrinsic<[llvm_float_ty], [llvm_float_ty]>;
 
-    def int_nvvm_trunc_d : NVVMBuiltin,
-        DefaultAttrsIntrinsic<[llvm_double_ty], [llvm_double_ty]>;
-  }
+  def int_nvvm_trunc_d : NVVMBuiltin,
+      PureIntrinsic<[llvm_double_ty], [llvm_double_ty]>;
 
-//
-// Saturate
-//
-  let IntrProperties = [IntrNoMem, IntrSpeculatable] in {
-    foreach ftz = ["", "_ftz"] in
-      def int_nvvm_saturate # ftz # _f : NVVMBuiltin,
-          DefaultAttrsIntrinsic<[llvm_float_ty], [llvm_float_ty]>;
+  //
+  // Saturate
+  //
+  foreach ftz = ["", "_ftz"] in
+    def int_nvvm_saturate # ftz # _f : NVVMBuiltin,
+        PureIntrinsic<[llvm_float_ty], [llvm_float_ty]>;
 
-    def int_nvvm_saturate_d : NVVMBuiltin,
-        DefaultAttrsIntrinsic<[llvm_double_ty], [llvm_double_ty]>;
-  }
+  def int_nvvm_saturate_d : NVVMBuiltin,
+      PureIntrinsic<[llvm_double_ty], [llvm_double_ty]>;
 
-//
-// Exp2  Log2
-//
+  //
+  // Exp2  Log2
+  //
   let IntrProperties = [IntrNoMem] in {
----------------
AlexMaclean wrote:

NVIDIA hasn't published information about the latency of these instructions, we assume here that they may be expensive and hence we don't want to speculatively execute them. If there were reliable public latency info we could implement a cost model and make these speculatable but this isn't in scope for this change. 

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