[llvm] r352568 - [WebAssembly] Lower SCALAR_TO_VECTOR to splats

Tue Jan 29 15:44:48 PST 2019

Author: tlively
Date: Tue Jan 29 15:44:48 2019
New Revision: 352568

URL: http://llvm.org/viewvc/llvm-project?rev=352568&view=rev
Log:
[WebAssembly] Lower SCALAR_TO_VECTOR to splats

Reviewers: aheejin

Subscribers: dschuff, sbc100, jgravelle-google, hiraditya, sunfish

Differential Revision: https://reviews.llvm.org/D57269

Added:
    llvm/trunk/test/CodeGen/WebAssembly/simd-scalar-to-vector.ll
Modified:
    llvm/trunk/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td

Modified: llvm/trunk/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td?rev=352568&r1=352567&r2=352568&view=diff
==============================================================================

--- llvm/trunk/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td (original)
+++ llvm/trunk/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td Tue Jan 29 15:44:48 2019
@@ -230,6 +230,19 @@ defm "" : Splat<v2i64, "i64x2", I64, spl
 defm "" : Splat<v4f32, "f32x4", F32, splat4, 18>;
 defm "" : Splat<v2f64, "f64x2", F64, splat2, 21>;
 
+// scalar_to_vector leaves high lanes undefined, so can be a splat
+class ScalarSplatPat<ValueType vec_t, ValueType lane_t,
+                     WebAssemblyRegClass reg_t> :
+  Pat<(vec_t (scalar_to_vector (lane_t reg_t:$x))),
+      (!cast<Instruction>("SPLAT_"#vec_t) reg_t:$x)>;
+
+def : ScalarSplatPat<v16i8, i32, I32>;
+def : ScalarSplatPat<v8i16, i32, I32>;
+def : ScalarSplatPat<v4i32, i32, I32>;
+def : ScalarSplatPat<v2i64, i64, I64>;
+def : ScalarSplatPat<v4f32, f32, F32>;
+def : ScalarSplatPat<v2f64, f64, F64>;
+
 //===----------------------------------------------------------------------===//
 // Accessing lanes
 //===----------------------------------------------------------------------===//

Added: llvm/trunk/test/CodeGen/WebAssembly/simd-scalar-to-vector.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/WebAssembly/simd-scalar-to-vector.ll?rev=352568&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/WebAssembly/simd-scalar-to-vector.ll (added)
+++ llvm/trunk/test/CodeGen/WebAssembly/simd-scalar-to-vector.ll Tue Jan 29 15:44:48 2019
@@ -0,0 +1,43 @@
+; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+simd128 | FileCheck %s
+
+; Test that scalar_to_vector is lowered into a splat correctly.
+; This bugpoint-reduced code turns into the selection dag below.
+; TODO: find small test cases that produce scalar_to_vector dag nodes
+; to make this test more readable and comprehensive.
+
+;   t0: ch = EntryToken
+; t32: i32,ch = load<(load 4 from `<2 x i16>* undef`, align 1)> t0, undef:i32, undef:i32
+;   t33: v4i32 = scalar_to_vector t32
+; t34: v8i16 = bitcast t33
+;       t51: i32 = extract_vector_elt t34, Constant:i32<0>
+;   t52: ch = store<(store 2 into `<4 x i16>* undef`, align 1), trunc to i16> t32:1, t51, undef:i32, undef:i32
+;       t50: i32 = extract_vector_elt t34, Constant:i32<1>
+;     t53: ch = store<(store 2 into `<4 x i16>* undef` + 2, align 1), trunc to i16> t32:1, t50, undef:i32, undef:i32
+;       t49: i32 = extract_vector_elt t34, Constant:i32<2>
+;     t55: ch = store<(store 2 into `<4 x i16>* undef` + 4, align 1), trunc to i16> t32:1, t49, undef:i32, undef:i32
+;       t48: i32 = extract_vector_elt t34, Constant:i32<3>
+;     t57: ch = store<(store 2 into `<4 x i16>* undef` + 6, align 1), trunc to i16> t32:1, t48, undef:i32, undef:i32
+;   t58: ch = TokenFactor t52, t53, t55, t57
+; t24: ch = WebAssemblyISD::RETURN t58
+
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; CHECK-LABEL: foo:
+; CHECK: i32x4.splat
+define void @foo() {
+entry:
+  %a = load <2 x i16>, <2 x i16>* undef, align 1
+  %b = shufflevector <2 x i16> %a, <2 x i16> undef, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %0 = bitcast <8 x i16> %b to <16 x i8>
+  %shuffle.i214 = shufflevector <16 x i8> %0, <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef>, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
+  %1 = bitcast <16 x i8> %shuffle.i214 to <8 x i16>
+  %add82 = add <8 x i16> %1, zeroinitializer
+  %2 = select <8 x i1> undef, <8 x i16> undef, <8 x i16> %add82
+  %3 = bitcast <8 x i16> %2 to <16 x i8>
+  %shuffle.i204 = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+  %4 = bitcast <16 x i8> %shuffle.i204 to <8 x i16>
+  %dst2.0.vec.extract = shufflevector <8 x i16> %4, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  store <4 x i16> %dst2.0.vec.extract, <4 x i16>* undef, align 1
+  ret void
+}


