maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[ARM,MVE] Use VMOV.{S8,S16} for sign-extended extractelement. 

MVE includes instructions that extract an 8- or 16-bit lane from a
vector and sign-extend it into the output 32-bit GPR. `ARMInstrMVE.td`
already included isel patterns to select those instructions in
response to the `ARMISD::VGETLANEs` selection-DAG node type. But
`ARMISD::VGETLANEs` was never actually generated, because the code
that creates it was conditioned on NEON only.

It's an easy fix to enable the same code for integer MVE, and now IR
that sign-extends the result of an extractelement (whether explicitly
or as part of the function call ABI) will use `vmov.s8` instead of
`vmov.u8` followed by `sxtb`.

Reviewers: SjoerdMeijer, dmgreen, ostannard

Subscribers: kristof.beyls, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70132
This commit is contained in:
Simon Tatham 2019-11-12 14:48:22 +00:00
parent 1d55c9e59e
commit 5b9e4daef0
2 changed files with 90 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
llvm/lib/Target/ARM/ARMISelLowering.cpp
View File

@ -13843,11 +13843,12 @@ static SDValue PerformExtendCombine(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
SDValue N0 = N->getOperand(0);
// Check for sign- and zero-extensions of vector extract operations of 8-
// and 16-bit vector elements. NEON supports these directly. They are
// Check for sign- and zero-extensions of vector extract operations of 8- and
// 16-bit vector elements. NEON and MVE support these directly. They are
// handled during DAG combining because type legalization will promote them
// to 32-bit types and it is messy to recognize the operations after that.
if (ST->hasNEON() && N0.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
if ((ST->hasNEON() || ST->hasMVEIntegerOps()) &&
N0.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
SDValue Vec = N0.getOperand(0);
SDValue Lane = N0.getOperand(1);
EVT VT = N->getValueType(0);

86
llvm/test/CodeGen/Thumb2/mve-extractelt.ll Normal file
View File

@ -0,0 +1,86 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s
define arm_aapcs_vfpcc i32 @u8_explicit_extend(<16 x i8> %a) {
; CHECK-LABEL: u8_explicit_extend:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmov.u8 r0, q0[10]
; CHECK-NEXT: bx lr
entry:
%0 = extractelement <16 x i8> %a, i32 10
%1 = zext i8 %0 to i32
ret i32 %1
}
define arm_aapcs_vfpcc i32 @s8_explicit_extend(<16 x i8> %a) {
; CHECK-LABEL: s8_explicit_extend:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmov.s8 r0, q0[10]
; CHECK-NEXT: bx lr
entry:
%0 = extractelement <16 x i8> %a, i32 10
%1 = sext i8 %0 to i32
ret i32 %1
}
define arm_aapcs_vfpcc i8 @u8_extend_via_pcs(<16 x i8> %a) {
; CHECK-LABEL: u8_extend_via_pcs:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmov.u8 r0, q0[10]
; CHECK-NEXT: bx lr
entry:
%0 = extractelement <16 x i8> %a, i32 10
ret i8 %0
}
define arm_aapcs_vfpcc signext i8 @s8_extend_via_pcs(<16 x i8> %a) {
; CHECK-LABEL: s8_extend_via_pcs:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmov.s8 r0, q0[10]
; CHECK-NEXT: bx lr
entry:
%0 = extractelement <16 x i8> %a, i32 10
ret i8 %0
}
define arm_aapcs_vfpcc i32 @u16_explicit_extend(<8 x i16> %a) {
; CHECK-LABEL: u16_explicit_extend:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmov.u16 r0, q0[5]
; CHECK-NEXT: bx lr
entry:
%0 = extractelement <8 x i16> %a, i32 5
%1 = zext i16 %0 to i32
ret i32 %1
}
define arm_aapcs_vfpcc i32 @s16_explicit_extend(<8 x i16> %a) {
; CHECK-LABEL: s16_explicit_extend:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmov.s16 r0, q0[5]
; CHECK-NEXT: bx lr
entry:
%0 = extractelement <8 x i16> %a, i32 5
%1 = sext i16 %0 to i32
ret i32 %1
}
define arm_aapcs_vfpcc i16 @u16_extend_via_pcs(<8 x i16> %a) {
; CHECK-LABEL: u16_extend_via_pcs:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmov.u16 r0, q0[5]
; CHECK-NEXT: bx lr
entry:
%0 = extractelement <8 x i16> %a, i32 5
ret i16 %0
}
define arm_aapcs_vfpcc signext i16 @s16_extend_via_pcs(<8 x i16> %a) {
; CHECK-LABEL: s16_extend_via_pcs:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmov.s16 r0, q0[5]
; CHECK-NEXT: bx lr
entry:
%0 = extractelement <8 x i16> %a, i32 5
ret i16 %0
}