[mlir][arith] Add shift integration tests

[pingshiyu]

Shift left/right regression tests, from the original larger PR that has been broken down: #92272

[llvmbot]

@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-mlir-arith

Author: Jacob Yu (pingshiyu)

Changes

Shift left/right regression tests, from the original larger PR that has been broken down: #92272

---

Full diff: https://github.com/llvm/llvm-project/pull/98183.diff

1 Files Affected:

• (added) mlir/test/Integration/Dialect/Arith/CPU/shifts.mlir (+160)

diff --git a/mlir/test/Integration/Dialect/Arith/CPU/shifts.mlir b/mlir/test/Integration/Dialect/Arith/CPU/shifts.mlir
new file mode 100644
index 0000000000000..519ecaf210812
--- /dev/null
+++ b/mlir/test/Integration/Dialect/Arith/CPU/shifts.mlir
@@ -0,0 +1,160 @@
+// RUN: mlir-opt %s --test-lower-to-llvm | \
+// RUN:   mlir-cpu-runner -e entry -entry-point-result=void \
+// RUN:                   --shared-libs=%mlir_c_runner_utils | \
+// RUN:   FileCheck %s --match-full-lines
+
+func.func @shrsi_i8(%v1 : i8, %v2 : i8) {
+  vector.print str "@shrsi_i8\n"
+  %res = arith.shrsi %v1, %v2 : i8
+  vector.print %res : i8
+  return
+}
+
+func.func @shrui_i8(%v1 : i8, %v2 : i8) {
+  vector.print str "@shrui_i8\n"
+  %res = arith.shrui %v1, %v2 : i8
+  vector.print %res : i8
+  return
+}
+
+func.func @shli_i8(%v1 : i8, %v2 : i8) {
+  vector.print str "@shli_i8\n"
+  %res = arith.shli %v1, %v2 : i8
+  vector.print %res : i8
+  return
+}
+
+func.func @shrsi_i1(%v1 : i1, %v2 : i1) {
+  vector.print str "@shrsi_i1\n"
+  %res = arith.shrsi %v1, %v2 : i1
+  vector.print %res : i1
+  return
+}
+
+func.func @shrui_i1(%v1 : i1, %v2 : i1) {
+  vector.print str "@shrui_i1\n"
+  %res = arith.shrui %v1, %v2 : i1
+  vector.print %res : i1
+  return
+}
+
+func.func @shli_i1(%v1 : i1, %v2 : i1) {
+  vector.print str "@shli_i1\n"
+  %res = arith.shli %v1, %v2 : i1
+  vector.print %res : i1
+  return
+}
+
+func.func @shrsi() {
+    // ------------------------------------------------
+    // Test i1
+    // ------------------------------------------------
+    %false = arith.constant 0 : i1
+
+    // shift by zero : i1 should be non poison
+    // shrsi 0 0 : i1 = 0
+    // CHECK-LABEL: @shrsi_i1
+    // CHECK-NEXT:  0
+    func.call @shrsi_i1(%false, %false) : (i1, i1) -> ()
+
+    // ------------------------------------------------
+    // Test i8
+    // ------------------------------------------------
+    %c7 = arith.constant 7 : i8
+    %cn10 = arith.constant -10 : i8
+    %c0 = arith.constant 0 : i8
+
+    // shrsi preserves signs
+    // shrsi -10 7 : i8 = -1
+    // CHECK-LABEL: @shrsi_i8
+    // CHECK-NEXT:  -1
+    func.call @shrsi_i8(%cn10, %c7) : (i8, i8) -> ()
+
+    // shift on zero is identity
+    // shrsi 7 0 : i8 = 7
+    // CHECK-LABEL: @shrsi_i8
+    // CHECK-NEXT:  7
+    func.call @shrsi_i8(%c7, %c0) : (i8, i8) -> ()
+
+    // ------------------------------------------------
+    // TODO: Test i16, i32 etc..
+    // ------------------------------------------------
+
+    return
+}
+
+func.func @shrui() {
+    // ------------------------------------------------
+    // Test i1
+    // ------------------------------------------------
+    %false = arith.constant 0 : i1
+
+    // shift by zero : i1 should be non poison
+    // shrui 0 0 : i1 = 0
+    // CHECK-LABEL: @shrui_i1
+    // CHECK-NEXT:  0
+    func.call @shrui_i1(%false, %false) : (i1, i1) -> ()
+
+    // ------------------------------------------------
+    // Test i8
+    // ------------------------------------------------
+    %cn10 = arith.constant -10 : i8
+    %c0 = arith.constant 0 : i8
+
+    // shift on zero is identity
+    // shrsi -10 0 : i8 = -10
+    // CHECK-LABEL: @shrui_i8
+    // CHECK-NEXT:  -10
+    func.call @shrui_i8(%cn10, %c0) : (i8, i8) -> ()
+
+    // ------------------------------------------------
+    // TODO: Test i16, i32 etc..
+    // ------------------------------------------------
+
+    return
+}
+
+func.func @shli() {
+    // ------------------------------------------------
+    // Test i1
+    // ------------------------------------------------
+    %false = arith.constant 0 : i1
+
+    // shift by zero : i1 should be non poison
+    // shli 0 0 : i1 = 0
+    // CHECK-LABEL: @shli_i1
+    // CHECK-NEXT:  0
+    func.call @shli_i1(%false, %false) : (i1, i1) -> ()
+
+    // ------------------------------------------------
+    // Test i8
+    // ------------------------------------------------
+    %c7 = arith.constant 7 : i8
+    %c0 = arith.constant 0 : i8
+    %cn100 = arith.constant -100 : i8
+
+    // shift on zero is identity
+    // shli 7 0 : i8 = 7
+    // CHECK-LABEL: @shli_i8
+    // CHECK-NEXT:  7
+    func.call @shli_i8(%c7, %c0) : (i8, i8) -> ()
+
+    // shli on i8, value goes off into the void (overflow/modulus needed)
+    // shli (-100), 7
+    // CHECK-LABEL: @shli_i8
+    // CHECK-NEXT:  0
+    func.call @shli_i8(%cn100, %c7) : (i8, i8) -> ()
+
+    // ------------------------------------------------
+    // TODO: Test i16, i32 etc..
+    // ------------------------------------------------
+
+    return
+}
+
+func.func @entry() {
+    func.call @shrsi() : () -> ()
+    func.call @shrui() : () -> ()
+    func.call @shli() : () -> ()
+    return
+}

[kuhar]

We cannot check for poison in integration tests... I'd suggest removing this comment

[kuhar]

same here

[kuhar]

also here

[kuhar]

I don't understand this comment

[joker-eph]

Same as in the other PR, I'd like to see more motivation for these tests. I'm not sure about the whole value they add (we don't need to have e2e for everything like this!) and a such the description should be fleshed out to make it more clear.

[banach-space]

Given the discussion here:

• [mlir][arith] Add more canonicalization and integration tests coverage #92272,

and the resulting update to the MLIR guidelines on e2e tests:

• Add a paragraph on integration tests mlir-www#203,

I suggest that we park this for now.

@pingshiyu, thank you so much for your contribution and for bearing with us throughout this process. Apologies for not resolving the earlier discussion sooner—it would have saved you a lot of effort. I hope this doesn’t discourage you from contributing to MLIR in the future, and we truly appreciate your work!