[llvm] [WebAssembly] Support shuffle for F16x8 vectors. (PR #127857)

[Brendan Dahl via llvm-commits]

https://github.com/brendandahl created https://github.com/llvm/llvm-project/pull/127857

None

>From b9e527c47c690c8c9bc9175f81c3975b7156c629 Mon Sep 17 00:00:00 2001
From: Brendan Dahl <brendan.dahl at gmail.com>
Date: Wed, 19 Feb 2025 19:27:39 +0000
Subject: [PATCH] [WebAssembly] Support shuffle for F16x8 vectors.

---
 .../WebAssembly/WebAssemblyISelLowering.cpp   |  3 +++
 .../WebAssembly/WebAssemblyInstrSIMD.td       |  2 +-
 .../CodeGen/WebAssembly/half-precision.ll     | 24 +++++++++++++++++++
 3 files changed, 28 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index fedad25c775e2..2877e909933fe 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
@@ -228,6 +228,9 @@ WebAssemblyTargetLowering::WebAssemblyTargetLowering(
                    MVT::v2f64})
       setOperationAction(ISD::VECTOR_SHUFFLE, T, Custom);
 
+    if (Subtarget->hasFP16())
+      setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v8f16, Custom);
+
     // Support splatting
     for (auto T : {MVT::v16i8, MVT::v8i16, MVT::v4i32, MVT::v4f32, MVT::v2i64,
                    MVT::v2f64})
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
index 14acc623ce24d..c591e5ef181a4 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
@@ -558,7 +558,7 @@ defm SHUFFLE :
 // Shuffles after custom lowering
 def wasm_shuffle_t : SDTypeProfile<1, 18, []>;
 def wasm_shuffle : SDNode<"WebAssemblyISD::SHUFFLE", wasm_shuffle_t>;
-foreach vec = StdVecs in {
+foreach vec = AllVecs in {
 // The @llvm.wasm.shuffle intrinsic has immediate arguments that become TargetConstants.
 def : Pat<(vec.vt (wasm_shuffle (vec.vt V128:$x), (vec.vt V128:$y),
             (i32 timm:$m0), (i32 timm:$m1),
diff --git a/llvm/test/CodeGen/WebAssembly/half-precision.ll b/llvm/test/CodeGen/WebAssembly/half-precision.ll
index 5f0ba4aa9c3c4..c300619ac1acc 100644
--- a/llvm/test/CodeGen/WebAssembly/half-precision.ll
+++ b/llvm/test/CodeGen/WebAssembly/half-precision.ll
@@ -335,3 +335,27 @@ define void @store_v8f16(<8 x half> %v, ptr %p) {
   store <8 x half> %v , ptr %p
   ret void
 }
+
+; ==============================================================================
+; Shuffle
+; ==============================================================================
+define <8 x half> @shuffle_v8f16(<8 x half> %x, <8 x half> %y) {
+; CHECK-LABEL: shuffle_v8f16:
+; CHECK:         .functype shuffle_v8f16 (v128, v128) -> (v128)
+; CHECK-NEXT:    i8x16.shuffle $push0=, $0, $1, 0, 1, 18, 19, 4, 5, 22, 23, 8, 9, 26, 27, 12, 13, 30, 31
+; CHECK-NEXT:    return $pop0
+  %res = shufflevector <8 x half> %x, <8 x half> %y,
+    <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x half> %res
+}
+
+define <8 x half> @shuffle_undef_v8f16(<8 x half> %x, <8 x half> %y) {
+; CHECK-LABEL: shuffle_undef_v8f16:
+; CHECK:         .functype shuffle_undef_v8f16 (v128, v128) -> (v128)
+; CHECK-NEXT:    i8x16.shuffle $push0=, $0, $0, 2, 3, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1
+; CHECK-NEXT:    return $pop0
+  %res = shufflevector <8 x half> %x, <8 x half> %y,
+    <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef,
+               i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <8 x half> %res
+}