[Mlir-commits] [mlir] [MLIR][NVVM] Enable nvvm intrinsics import (PR #68843)

[Tobias Gysi]

================
@@ -0,0 +1,430 @@
+; RUN: mlir-translate -import-llvm %s | FileCheck %s
+
+; CHECK-LABEL: @nvvm_special_regs
+define i32 @nvvm_special_regs() {
+  ; CHECK: = nvvm.read.ptx.sreg.tid.x : i32
+  %1 = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+  ; CHECK: = nvvm.read.ptx.sreg.tid.y : i32
+  %2 = call i32 @llvm.nvvm.read.ptx.sreg.tid.y()
+  ; CHECK: = nvvm.read.ptx.sreg.tid.z : i32
+  %3 = call i32 @llvm.nvvm.read.ptx.sreg.tid.z()
+  ; CHECK: = nvvm.read.ptx.sreg.ntid.x : i32
+  %4 = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
+  ; CHECK: = nvvm.read.ptx.sreg.ntid.y : i32
+  %5 = call i32 @llvm.nvvm.read.ptx.sreg.ntid.y()
+  ; CHECK: = nvvm.read.ptx.sreg.ntid.z : i32
+  %6 = call i32 @llvm.nvvm.read.ptx.sreg.ntid.z()
+  ; CHECK: = nvvm.read.ptx.sreg.ctaid.x : i32
+  %7 = call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x()
+  ; CHECK: = nvvm.read.ptx.sreg.ctaid.y : i32
+  %8 = call i32 @llvm.nvvm.read.ptx.sreg.ctaid.y()
+  ; CHECK: = nvvm.read.ptx.sreg.ctaid.z : i32
+  %9 = call i32 @llvm.nvvm.read.ptx.sreg.ctaid.z()
+  ; CHECK: = nvvm.read.ptx.sreg.nctaid.x : i32
+  %10 = call i32 @llvm.nvvm.read.ptx.sreg.nctaid.x()
+  ; CHECK: = nvvm.read.ptx.sreg.nctaid.y : i32
+  %11 = call i32 @llvm.nvvm.read.ptx.sreg.nctaid.y()
+  ; CHECK: = nvvm.read.ptx.sreg.nctaid.z : i32
+  %12 = call i32 @llvm.nvvm.read.ptx.sreg.nctaid.z()
+  ; CHECK: = nvvm.read.ptx.sreg.warpsize : i32
+  %13 = call i32 @llvm.nvvm.read.ptx.sreg.warpsize()
+  ; CHECK: = nvvm.read.ptx.sreg.laneid : i32
+  %14 = call i32 @llvm.nvvm.read.ptx.sreg.laneid()
+  ; CHECK: = nvvm.read.ptx.sreg.clusterid.x : i32
+  %15 = call i32 @llvm.nvvm.read.ptx.sreg.clusterid.x()
+  ; CHECK: = nvvm.read.ptx.sreg.clusterid.y : i32
+  %16 = call i32 @llvm.nvvm.read.ptx.sreg.clusterid.y()
+  ; CHECK: = nvvm.read.ptx.sreg.clusterid.z : i32
+  %17 = call i32 @llvm.nvvm.read.ptx.sreg.clusterid.z()
+  ; CHECK: = nvvm.read.ptx.sreg.nclusterid.x : i32
+  %18 = call i32 @llvm.nvvm.read.ptx.sreg.nclusterid.x()
+  ; CHECK: = nvvm.read.ptx.sreg.nclusterid.y : i32
+  %19 = call i32 @llvm.nvvm.read.ptx.sreg.nclusterid.y()
+  ; CHECK: = nvvm.read.ptx.sreg.nclusterid.z : i32
+  %20 = call i32 @llvm.nvvm.read.ptx.sreg.nclusterid.z()
+  ; CHECK: = nvvm.read.ptx.sreg.cluster.ctaid.x : i32
+  %21 = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.x()
+  ; CHECK: = nvvm.read.ptx.sreg.cluster.ctaid.y : i32
+  %22 = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.y()
+  ; CHECK: = nvvm.read.ptx.sreg.cluster.ctaid.z : i32
+  %23 = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.z()
+  ; CHECK: = nvvm.read.ptx.sreg.cluster.nctaid.x : i32
+  %24 = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()
+  ; CHECK: = nvvm.read.ptx.sreg.cluster.nctaid.y : i32
+  %25 = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.y()
+  ; CHECK: = nvvm.read.ptx.sreg.cluster.nctaid.z : i32
+  %26 = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.z()
+  ; CHECK: = nvvm.read.ptx.sreg.cluster.ctarank : i32
+  %27 = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctarank()
+  ; CHECK: = nvvm.read.ptx.sreg.cluster.nctarank : i32
+  %28 = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctarank()
+  ret i32 %1
+}
+
+; CHECK-LABEL: @nvvm_rcp
+define float @nvvm_rcp(float %0) {
+  ; CHECK: = nvvm.rcp.approx.ftz.f %{{.*}} : f32
+  %2 = call float @llvm.nvvm.rcp.approx.ftz.f(float %0)
+  ret float %2
+}
+
+; TODO Below intrinsics not yet supported
----------------
gysit wrote:

```suggestion
; TODO: Support the intrinsics below once they derive from NVVM_IntrOp rather than from NVVM_Op.
```
nit: Can you mention the reason here (at least this is what I understood from your comment on the PR.

https://github.com/llvm/llvm-project/pull/68843