[llvm-commits] [llvm] r135313 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/avx-256-arith.s
Bruno Cardoso Lopes
bruno.cardoso at gmail.com
Fri Jul 15 15:24:33 PDT 2011
Author: bruno
Date: Fri Jul 15 17:24:33 2011
New Revision: 135313
URL: http://llvm.org/viewvc/llvm-project?rev=135313&view=rev
Log:
Fix a couple of things:
1) Make non-legal 256-bit loads to be promoted to v4i64. This lets us
canonize the loads and handle things the same way we use to handle
for 128-bit registers. Despite of what one of the removed comments
explained, the load promotion would not mess with VPERM, it's only a
matter of doing the appropriate bitcasts when this instructions comes
to be introduced. Also make LOAD v8i32 legal.
2) Doing 1) exposed two bugs:
- v4i64 was being promoted to itself for several opcodes (introduced
in r124447 by David Greene) causing endless recursion and the stack to
explode.
- there was no support for allOnes BUILD_VECTORs and ANDNP would fail to
match because it was generating early target constant pools during
lowering.
3) The testcases are already checked-in, doing 1) exposed the
bugs in the current testcases.
4) Tidy up code to be more clear and explicit about AVX.
Added:
llvm/trunk/test/CodeGen/X86/avx-256-arith.s
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=135313&r1=135312&r2=135313&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Fri Jul 15 17:24:33 2011
@@ -988,7 +988,6 @@
addRegisterClass(MVT::v32i8, X86::VR256RegisterClass);
setOperationAction(ISD::LOAD, MVT::v8f32, Legal);
- setOperationAction(ISD::LOAD, MVT::v8i32, Legal);
setOperationAction(ISD::LOAD, MVT::v4f64, Legal);
setOperationAction(ISD::LOAD, MVT::v4i64, Legal);
@@ -1006,64 +1005,51 @@
setOperationAction(ISD::FSQRT, MVT::v4f64, Legal);
setOperationAction(ISD::FNEG, MVT::v4f64, Custom);
- // Custom lower build_vector, vector_shuffle, scalar_to_vector,
- // insert_vector_elt extract_subvector and extract_vector_elt for
- // 256-bit types.
+ // Custom lower several nodes for 256-bit types.
for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
- i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE;
- ++i) {
- MVT::SimpleValueType VT = (MVT::SimpleValueType)i;
- // Do not attempt to custom lower non-256-bit vectors
- if (!isPowerOf2_32(MVT(VT).getVectorNumElements())
- || (MVT(VT).getSizeInBits() < 256))
- continue;
- setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
- setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
- setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
- setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
- setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Custom);
- }
- // Custom-lower insert_subvector and extract_subvector based on
- // the result type.
- for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
- i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE;
- ++i) {
- MVT::SimpleValueType VT = (MVT::SimpleValueType)i;
- // Do not attempt to custom lower non-256-bit vectors
- if (!isPowerOf2_32(MVT(VT).getVectorNumElements()))
+ i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) {
+ MVT::SimpleValueType SVT = (MVT::SimpleValueType)i;
+ EVT VT = SVT;
+
+ // Extract subvector is special because the value type
+ // (result) is 128-bit but the source is 256-bit wide.
+ if (VT.is128BitVector())
+ setOperationAction(ISD::EXTRACT_SUBVECTOR, SVT, Custom);
+
+ // Do not attempt to custom lower other non-256-bit vectors
+ if (!VT.is256BitVector())
continue;
- if (MVT(VT).getSizeInBits() == 128) {
- setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Custom);
- }
- else if (MVT(VT).getSizeInBits() == 256) {
- setOperationAction(ISD::INSERT_SUBVECTOR, VT, Custom);
- }
+ setOperationAction(ISD::BUILD_VECTOR, SVT, Custom);
+ setOperationAction(ISD::VECTOR_SHUFFLE, SVT, Custom);
+ setOperationAction(ISD::INSERT_VECTOR_ELT, SVT, Custom);
+ setOperationAction(ISD::EXTRACT_VECTOR_ELT, SVT, Custom);
+ setOperationAction(ISD::SCALAR_TO_VECTOR, SVT, Custom);
+ setOperationAction(ISD::INSERT_SUBVECTOR, SVT, Custom);
}
// Promote v32i8, v16i16, v8i32 select, and, or, xor to v4i64.
- // Don't promote loads because we need them for VPERM vector index versions.
+ for (unsigned i = (unsigned)MVT::v32i8; i != (unsigned)MVT::v4i64; ++i) {
+ MVT::SimpleValueType SVT = (MVT::SimpleValueType)i;
+ EVT VT = SVT;
- for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
- VT != (unsigned)MVT::LAST_VECTOR_VALUETYPE;
- VT++) {
- if (!isPowerOf2_32(MVT((MVT::SimpleValueType)VT).getVectorNumElements())
- || (MVT((MVT::SimpleValueType)VT).getSizeInBits() < 256))
+ // Do not attempt to promote non-256-bit vectors
+ if (!VT.is256BitVector())
continue;
- setOperationAction(ISD::AND, (MVT::SimpleValueType)VT, Promote);
- AddPromotedToType (ISD::AND, (MVT::SimpleValueType)VT, MVT::v4i64);
- setOperationAction(ISD::OR, (MVT::SimpleValueType)VT, Promote);
- AddPromotedToType (ISD::OR, (MVT::SimpleValueType)VT, MVT::v4i64);
- setOperationAction(ISD::XOR, (MVT::SimpleValueType)VT, Promote);
- AddPromotedToType (ISD::XOR, (MVT::SimpleValueType)VT, MVT::v4i64);
- //setOperationAction(ISD::LOAD, (MVT::SimpleValueType)VT, Promote);
- //AddPromotedToType (ISD::LOAD, (MVT::SimpleValueType)VT, MVT::v4i64);
- setOperationAction(ISD::SELECT, (MVT::SimpleValueType)VT, Promote);
- AddPromotedToType (ISD::SELECT, (MVT::SimpleValueType)VT, MVT::v4i64);
+
+ setOperationAction(ISD::AND, SVT, Promote);
+ AddPromotedToType (ISD::AND, SVT, MVT::v4i64);
+ setOperationAction(ISD::OR, SVT, Promote);
+ AddPromotedToType (ISD::OR, SVT, MVT::v4i64);
+ setOperationAction(ISD::XOR, SVT, Promote);
+ AddPromotedToType (ISD::XOR, SVT, MVT::v4i64);
+ setOperationAction(ISD::LOAD, SVT, Promote);
+ AddPromotedToType (ISD::LOAD, SVT, MVT::v4i64);
+ setOperationAction(ISD::SELECT, SVT, Promote);
+ AddPromotedToType (ISD::SELECT, SVT, MVT::v4i64);
}
}
-
// SIGN_EXTEND_INREGs are evaluated by the extend type. Handle the expansion
// of this type with custom code.
for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
@@ -3852,19 +3838,24 @@
}
/// getOnesVector - Returns a vector of specified type with all bits set.
-///
+/// Always build ones vectors as <4 x i32> or <8 x i32> bitcasted to
+/// their original type, ensuring they get CSE'd.
static SDValue getOnesVector(EVT VT, SelectionDAG &DAG, DebugLoc dl) {
assert(VT.isVector() && "Expected a vector type");
+ assert((VT.is128BitVector() || VT.is256BitVector())
+ && "Expected a 128-bit or 256-bit vector type");
- // Always build ones vectors as <4 x i32> or <2 x i32> bitcasted to their dest
- // type. This ensures they get CSE'd.
SDValue Cst = DAG.getTargetConstant(~0U, MVT::i32);
+
SDValue Vec;
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
+ if (VT.is256BitVector()) {
+ SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
+ Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops, 8);
+ } else
+ Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
return DAG.getNode(ISD::BITCAST, dl, VT, Vec);
}
-
/// NormalizeMask - V2 is a splat, modify the mask (if needed) so all elements
/// that point to V2 points to its first element.
static SDValue NormalizeMask(ShuffleVectorSDNode *SVOp, SelectionDAG &DAG) {
@@ -4479,17 +4470,17 @@
return ConcatVectors(Lower, Upper, DAG);
}
- // All zero's are handled with pxor in SSE2 and above, xorps in SSE1.
- // All one's are handled with pcmpeqd. In AVX, zero's are handled with
- // vpxor in 128-bit and xor{pd,ps} in 256-bit, but no 256 version of pcmpeqd
- // is present, so AllOnes is ignored.
+ // All zero's:
+ // - pxor (SSE2), xorps (SSE1), vpxor (128 AVX), xorp[s|d] (256 AVX)
+ // All one's:
+ // - pcmpeqd (SSE2 and 128 AVX), fallback to constant pools (256 AVX)
if (ISD::isBuildVectorAllZeros(Op.getNode()) ||
- (Op.getValueType().getSizeInBits() != 256 &&
- ISD::isBuildVectorAllOnes(Op.getNode()))) {
- // Canonicalize this to <4 x i32> (SSE) to
+ ISD::isBuildVectorAllOnes(Op.getNode())) {
+ // Canonicalize this to <4 x i32> or <8 x 32> (SSE) to
// 1) ensure the zero vectors are CSE'd, and 2) ensure that i64 scalars are
// eliminated on x86-32 hosts.
- if (Op.getValueType() == MVT::v4i32)
+ if (Op.getValueType() == MVT::v4i32 ||
+ Op.getValueType() == MVT::v8i32)
return Op;
if (ISD::isBuildVectorAllOnes(Op.getNode()))
Added: llvm/trunk/test/CodeGen/X86/avx-256-arith.s
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx-256-arith.s?rev=135313&view=auto
==============================================================================
(empty)
More information about the llvm-commits
mailing list