[WebAssembly] Optimize away mask of 63 for sra and srl( zext (and i32 63)))

[badumbatish]

Follow up to #71844 after shl implementation

[llvmbot]

@llvm/pr-subscribers-backend-webassembly

Author: Jasmine Tang (badumbatish)

Changes

Follow up to #71844 after shl implementation

---

Full diff: https://github.com/llvm/llvm-project/pull/170128.diff

2 Files Affected:

• (modified) llvm/lib/Target/WebAssembly/WebAssemblyInstrInteger.td (+4)
• (modified) llvm/test/CodeGen/WebAssembly/masked-shifts.ll (+31)

diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyInstrInteger.td b/llvm/lib/Target/WebAssembly/WebAssemblyInstrInteger.td
index eb692679f5971..991507e883f28 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyInstrInteger.td
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyInstrInteger.td
@@ -109,6 +109,10 @@ def : Pat<(rotr I64:$lhs, (and I64:$rhs, 63)), (ROTR_I64 I64:$lhs, I64:$rhs)>;
 
 def : Pat<(shl I64:$lhs, (zext (and I32:$rhs, 63))),
                                (SHL_I64 I64:$lhs, (I64_EXTEND_U_I32 I32:$rhs))>;
+def : Pat<(sra I64:$lhs, (zext (and I32:$rhs, 63))),
+                               (SHR_S_I64 I64:$lhs, (I64_EXTEND_U_I32 I32:$rhs))>;
+def : Pat<(srl I64:$lhs, (zext (and I32:$rhs, 63))),
+                               (SHR_U_I64 I64:$lhs, (I64_EXTEND_U_I32 I32:$rhs))>;
 
 defm SELECT_I32 : I<(outs I32:$dst), (ins I32:$lhs, I32:$rhs, I32:$cond),
                     (outs), (ins),
diff --git a/llvm/test/CodeGen/WebAssembly/masked-shifts.ll b/llvm/test/CodeGen/WebAssembly/masked-shifts.ll
index 368f30fd5d7ed..95b0a8a866eb1 100644
--- a/llvm/test/CodeGen/WebAssembly/masked-shifts.ll
+++ b/llvm/test/CodeGen/WebAssembly/masked-shifts.ll
@@ -46,6 +46,21 @@ define i32 @sra_i32(i32 %v, i32 %x) {
   ret i32 %a
 }
 
+define i64 @sra_i64_zext(i64 %v, i32 %x) {
+; CHECK-LABEL: sra_i64_zext:
+; CHECK:         .functype sra_i64_zext (i64, i32) -> (i64)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    i64.extend_i32_u
+; CHECK-NEXT:    i64.shr_s
+; CHECK-NEXT:    # fallthrough-return
+  %m = and i32 %x, 63
+  %z = zext i32 %m to i64
+  %a = ashr i64 %v, %z
+  ret i64 %a
+}
+
 define i32 @srl_i32(i32 %v, i32 %x) {
 ; CHECK-LABEL: srl_i32:
 ; CHECK:         .functype srl_i32 (i32, i32) -> (i32)
@@ -59,6 +74,22 @@ define i32 @srl_i32(i32 %v, i32 %x) {
   ret i32 %a
 }
 
+define i64 @srl_i64_zext(i64 %v, i32 %x) {
+; CHECK-LABEL: srl_i64_zext:
+; CHECK:         .functype srl_i64_zext (i64, i32) -> (i64)
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    i64.extend_i32_u
+; CHECK-NEXT:    i64.shr_u
+; CHECK-NEXT:    # fallthrough-return
+  %m = and i32 %x, 63
+  %z = zext i32 %m to i64
+  %a = lshr i64 %v, %z
+  ret i64 %a
+}
+
+
 define i64 @shl_i64(i64 %v, i64 %x) {
 ; CHECK-LABEL: shl_i64:
 ; CHECK:         .functype shl_i64 (i64, i64) -> (i64)

[lukel97]

LGTM