[clang] 5e09e99 - [WebAssembly] Prototype extending pairwise add instructions
Thomas Lively via cfe-commits
cfe-commits at lists.llvm.org
Mon Dec 28 14:11:24 PST 2020
Author: Thomas Lively
Date: 2020-12-28T14:11:14-08:00
New Revision: 5e09e9979bc60f0fca0e80e7f72f1260bd1bbca5
URL: https://github.com/llvm/llvm-project/commit/5e09e9979bc60f0fca0e80e7f72f1260bd1bbca5
DIFF: https://github.com/llvm/llvm-project/commit/5e09e9979bc60f0fca0e80e7f72f1260bd1bbca5.diff
LOG: [WebAssembly] Prototype extending pairwise add instructions
As proposed in https://github.com/WebAssembly/simd/pull/380. This commit makes
the new instructions available only via clang builtins and LLVM intrinsics to
make their use opt-in while they are still being evaluated for inclusion in the
SIMD proposal.
Depends on D93771.
Differential Revision: https://reviews.llvm.org/D93775
Added:
Modified:
clang/include/clang/Basic/BuiltinsWebAssembly.def
clang/lib/CodeGen/CGBuiltin.cpp
clang/test/CodeGen/builtins-wasm.c
llvm/include/llvm/IR/IntrinsicsWebAssembly.td
llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll
llvm/test/MC/WebAssembly/simd-encodings.s
Removed:
################################################################################
diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index 023365bec7f8..d6860e0b13be 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -133,6 +133,12 @@ TARGET_BUILTIN(__builtin_wasm_extmul_high_i32x4_s_i64x2, "V2LLiV4iV4i", "nc", "s
TARGET_BUILTIN(__builtin_wasm_extmul_low_i32x4_u_i64x2, "V2ULLiV4UiV4Ui", "nc", "simd128")
TARGET_BUILTIN(__builtin_wasm_extmul_high_i32x4_u_i64x2, "V2ULLiV4UiV4Ui", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_extadd_pairwise_i8x16_s_i16x8, "V8sV16Sc", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_extadd_pairwise_i8x16_u_i16x8, "V8UsV16Uc", "nc", "simd128")
+
+TARGET_BUILTIN(__builtin_wasm_extadd_pairwise_i16x8_s_i32x4, "V4iV8s", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_extadd_pairwise_i16x8_u_i32x4, "V4UiV8Us", "nc", "simd128")
+
TARGET_BUILTIN(__builtin_wasm_bitselect, "V4iV4iV4iV4i", "nc", "simd128")
TARGET_BUILTIN(__builtin_wasm_signselect_i8x16, "V16ScV16ScV16ScV16Sc", "nc", "simd128")
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 60bfa90e22fc..0c02dbfe8469 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -16932,6 +16932,28 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType()));
return Builder.CreateCall(Callee, {LHS, RHS});
}
+ case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_s_i16x8:
+ case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_u_i16x8:
+ case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_s_i32x4:
+ case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_u_i32x4: {
+ Value *Vec = EmitScalarExpr(E->getArg(0));
+ unsigned IntNo;
+ switch (BuiltinID) {
+ case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_s_i16x8:
+ case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_s_i32x4:
+ IntNo = Intrinsic::wasm_extadd_pairwise_signed;
+ break;
+ case WebAssembly::BI__builtin_wasm_extadd_pairwise_i8x16_u_i16x8:
+ case WebAssembly::BI__builtin_wasm_extadd_pairwise_i16x8_u_i32x4:
+ IntNo = Intrinsic::wasm_extadd_pairwise_unsigned;
+ break;
+ default:
+ llvm_unreachable("unexptected builtin ID");
+ }
+
+ Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType()));
+ return Builder.CreateCall(Callee, Vec);
+ }
case WebAssembly::BI__builtin_wasm_bitselect: {
Value *V1 = EmitScalarExpr(E->getArg(0));
Value *V2 = EmitScalarExpr(E->getArg(1));
diff --git a/clang/test/CodeGen/builtins-wasm.c b/clang/test/CodeGen/builtins-wasm.c
index 76dd2622fe2f..a07c278c33af 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -603,6 +603,34 @@ u64x2 extmul_high_i32x4_u_i64x2(u32x4 x, u32x4 y) {
// WEBASSEMBLY-NEXT: ret
}
+i16x8 extadd_pairwise_i8x16_s_i16x8(i8x16 v) {
+ return __builtin_wasm_extadd_pairwise_i8x16_s_i16x8(v);
+ // WEBASSEMBLY: call <8 x i16> @llvm.wasm.extadd.pairwise.signed.v8i16(
+ // WEBASSEMBLY-SAME: <16 x i8> %v)
+ // WEBASSEMBLY-NEXT: ret
+}
+
+u16x8 extadd_pairwise_i8x16_u_i16x8(u8x16 v) {
+ return __builtin_wasm_extadd_pairwise_i8x16_u_i16x8(v);
+ // WEBASSEMBLY: call <8 x i16> @llvm.wasm.extadd.pairwise.unsigned.v8i16(
+ // WEBASSEMBLY-SAME: <16 x i8> %v)
+ // WEBASSEMBLY-NEXT: ret
+}
+
+i32x4 extadd_pairwise_i16x8_s_i32x4(i16x8 v) {
+ return __builtin_wasm_extadd_pairwise_i16x8_s_i32x4(v);
+ // WEBASSEMBLY: call <4 x i32> @llvm.wasm.extadd.pairwise.signed.v4i32(
+ // WEBASSEMBLY-SAME: <8 x i16> %v)
+ // WEBASSEMBLY-NEXT: ret
+}
+
+u32x4 extadd_pairwise_i16x8_u_i32x4(u16x8 v) {
+ return __builtin_wasm_extadd_pairwise_i16x8_u_i32x4(v);
+ // WEBASSEMBLY: call <4 x i32> @llvm.wasm.extadd.pairwise.unsigned.v4i32(
+ // WEBASSEMBLY-SAME: <8 x i16> %v)
+ // WEBASSEMBLY-NEXT: ret
+}
+
i32x4 dot_i16x8_s(i16x8 x, i16x8 y) {
return __builtin_wasm_dot_s_i32x4_i16x8(x, y);
// WEBASSEMBLY: call <4 x i32> @llvm.wasm.dot(<8 x i16> %x, <8 x i16> %y)
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index 0eede93792d6..d9a6aa78fdcd 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -290,6 +290,15 @@ def int_wasm_extmul_high_unsigned :
[LLVMSubdivide2VectorType<0>, LLVMSubdivide2VectorType<0>],
[IntrNoMem, IntrSpeculatable]>;
+def int_wasm_extadd_pairwise_signed :
+ Intrinsic<[llvm_anyvector_ty],
+ [LLVMSubdivide2VectorType<0>],
+ [IntrNoMem, IntrSpeculatable]>;
+def int_wasm_extadd_pairwise_unsigned :
+ Intrinsic<[llvm_anyvector_ty],
+ [LLVMSubdivide2VectorType<0>],
+ [IntrNoMem, IntrSpeculatable]>;
+
def int_wasm_signselect :
Intrinsic<[llvm_anyvector_ty],
[LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
index 707b7e3998d0..30e9fa21ba83 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
@@ -1246,6 +1246,17 @@ foreach t2 = !foldl(
) in
def : Pat<(t1 (bitconvert (t2 V128:$v))), (t1 V128:$v)>;
+// Extended pairwise addition
+defm "" : SIMDConvert<I16x8, I8x16, int_wasm_extadd_pairwise_signed,
+ "extadd_pairwise_i8x16_s", 0xc2>;
+defm "" : SIMDConvert<I16x8, I8x16, int_wasm_extadd_pairwise_unsigned,
+ "extadd_pairwise_i8x16_u", 0xc3>;
+defm "" : SIMDConvert<I32x4, I16x8, int_wasm_extadd_pairwise_signed,
+ "extadd_pairwise_i16x8_s", 0xa5>;
+defm "" : SIMDConvert<I32x4, I16x8, int_wasm_extadd_pairwise_unsigned,
+ "extadd_pairwise_i16x8_u", 0xa6>;
+
+
//===----------------------------------------------------------------------===//
// Quasi-Fused Multiply- Add and Subtract (QFMA/QFMS)
//===----------------------------------------------------------------------===//
diff --git a/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll b/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll
index da7343770de7..1475f2fd1b34 100644
--- a/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll
+++ b/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll
@@ -309,6 +309,26 @@ define <8 x i16> @extmul_high_u_v8i16(<16 x i8> %x, <16 x i8> %y) {
ret <8 x i16> %a
}
+; CHECK-LABEL: extadd_pairwise_s_v8i16:
+; SIMD128-NEXT: .functype extadd_pairwise_s_v8i16 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i16x8.extadd_pairwise_i8x16_s $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <8 x i16> @llvm.wasm.extadd.pairwise.signed.v8i16(<16 x i8>)
+define <8 x i16> @extadd_pairwise_s_v8i16(<16 x i8> %x) {
+ %a = call <8 x i16> @llvm.wasm.extadd.pairwise.signed.v8i16(<16 x i8> %x)
+ ret <8 x i16> %a
+}
+
+; CHECK-LABEL: extadd_pairwise_u_v8i16:
+; SIMD128-NEXT: .functype extadd_pairwise_u_v8i16 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i16x8.extadd_pairwise_i8x16_u $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <8 x i16> @llvm.wasm.extadd.pairwise.unsigned.v8i16(<16 x i8>)
+define <8 x i16> @extadd_pairwise_u_v8i16(<16 x i8> %x) {
+ %a = call <8 x i16> @llvm.wasm.extadd.pairwise.unsigned.v8i16(<16 x i8> %x)
+ ret <8 x i16> %a
+}
+
; CHECK-LABEL: any_v8i16:
; SIMD128-NEXT: .functype any_v8i16 (v128) -> (i32){{$}}
; SIMD128-NEXT: i16x8.any_true $push[[R:[0-9]+]]=, $0{{$}}
@@ -449,6 +469,27 @@ define <4 x i32> @extmul_high_u_v4i32(<8 x i16> %x, <8 x i16> %y) {
ret <4 x i32> %a
}
+; CHECK-LABEL: extadd_pairwise_s_v4i32:
+; SIMD128-NEXT: .functype extadd_pairwise_s_v4i32 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i32x4.extadd_pairwise_i16x8_s $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <4 x i32> @llvm.wasm.extadd.pairwise.signed.v4i32(<8 x i16>)
+define <4 x i32> @extadd_pairwise_s_v4i32(<8 x i16> %x) {
+ %a = call <4 x i32> @llvm.wasm.extadd.pairwise.signed.v4i32(<8 x i16> %x)
+ ret <4 x i32> %a
+}
+
+; CHECK-LABEL: extadd_pairwise_u_v4i32:
+; SIMD128-NEXT: .functype extadd_pairwise_u_v4i32 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i32x4.extadd_pairwise_i16x8_u $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <4 x i32> @llvm.wasm.extadd.pairwise.unsigned.v4i32(<8 x i16>)
+define <4 x i32> @extadd_pairwise_u_v4i32(<8 x i16> %x) {
+ %a = call <4 x i32> @llvm.wasm.extadd.pairwise.unsigned.v4i32(<8 x i16> %x)
+ ret <4 x i32> %a
+}
+
+
; CHECK-LABEL: any_v4i32:
; SIMD128-NEXT: .functype any_v4i32 (v128) -> (i32){{$}}
; SIMD128-NEXT: i32x4.any_true $push[[R:[0-9]+]]=, $0{{$}}
diff --git a/llvm/test/MC/WebAssembly/simd-encodings.s b/llvm/test/MC/WebAssembly/simd-encodings.s
index 91e1f07fe44c..ab3ce59f25df 100644
--- a/llvm/test/MC/WebAssembly/simd-encodings.s
+++ b/llvm/test/MC/WebAssembly/simd-encodings.s
@@ -724,4 +724,16 @@ main:
# CHECK: i64x2.signselect # encoding: [0xfd,0x94,0x01]
i64x2.signselect
+ # CHECK: i16x8.extadd_pairwise_i8x16_s # encoding: [0xfd,0xc2,0x01]
+ i16x8.extadd_pairwise_i8x16_s
+
+ # CHECK: i16x8.extadd_pairwise_i8x16_u # encoding: [0xfd,0xc3,0x01]
+ i16x8.extadd_pairwise_i8x16_u
+
+ # CHECK: i32x4.extadd_pairwise_i16x8_s # encoding: [0xfd,0xa5,0x01]
+ i32x4.extadd_pairwise_i16x8_s
+
+ # CHECK: i32x4.extadd_pairwise_i16x8_u # encoding: [0xfd,0xa6,0x01]
+ i32x4.extadd_pairwise_i16x8_u
+
end_function
More information about the cfe-commits
mailing list