This is an archive of the discontinued LLVM Phabricator instance.

Differential D90227

[VE] Add vector reduction instructions
ClosedPublic

Authored by kaz7 on Oct 27 2020, 5:14 AM.

Details

Reviewers
simoll
k-ishizaka
Commits
rGa65883a78a0c: [VE] Add vector reduction instructions
Summary

Add VSUMS/VSUMX/VFSUM/VMAXS/VMAXX/VFMAX/VRAND/VROR/VRXOR isntructions.
Add regression tests too.

Diff Detail

Unit TestsFailed

TimeTest
410 mslinux > HWAddressSanitizer-x86_64.TestCases::sizes.cpp

Event Timeline

kaz7 created this revision.Oct 27 2020, 5:14 AM
Herald added subscribers: llvm-commits, hiraditya. · View Herald TranscriptOct 27 2020, 5:14 AM
kaz7 requested review of this revision.Oct 27 2020, 5:14 AM
Harbormaster completed remote builds in B76537: Diff 300957.Oct 27 2020, 6:00 AM
simoll accepted this revision.Oct 27 2020, 8:40 AM
This revision is now accepted and ready to land.Oct 27 2020, 8:40 AM
This revision was landed with ongoing or failed builds.Oct 27 2020, 10:33 AM
Closed by commit rGa65883a78a0c: [VE] Add vector reduction instructions (authored by kaz7). · Explain Why
This revision was automatically updated to reflect the committed changes.
kaz7 added a commit: rGa65883a78a0c: [VE] Add vector reduction instructions.

Revision Contents

PathSize
llvm/
lib/
Target/
VE/
70 lines
test/
MC/
VE/
52 lines
28 lines
52 lines
20 lines
20 lines
20 lines
28 lines
28 lines

Diff 300957

llvm/lib/Target/VE/VEInstrVec.td

Show First 20 Lines • Show All 1,110 Lines • ▼ Show 20 Lines
// Section 8.13.17 - VFLTX (Vector Convert to Floating Pointer) // Section 8.13.17 - VFLTX (Vector Convert to Floating Pointer)
defm VCVTDL : RVF1m<"vcvt.d.l", 0xb8, V64, VM>; defm VCVTDL : RVF1m<"vcvt.d.l", 0xb8, V64, VM>;
// Section 8.13.18 - VCVS (Vector Convert to Single-format) // Section 8.13.18 - VCVS (Vector Convert to Single-format)
defm VCVTSD : RVF1m<"vcvt.s.d", 0x9f, V64, VM>; defm VCVTSD : RVF1m<"vcvt.s.d", 0x9f, V64, VM>;
// Section 8.13.19 - VCVD (Vector Convert to Double-format) // Section 8.13.19 - VCVD (Vector Convert to Double-format)
defm VCVTDS : RVF1m<"vcvt.d.s", 0x8f, V64, VM>; defm VCVTDS : RVF1m<"vcvt.d.s", 0x8f, V64, VM>;
//-----------------------------------------------------------------------------
// Section 8.14 - Vector Reduction Instructions
//-----------------------------------------------------------------------------
// Section 8.14.1 - VSUMS (Vector Sum Single)
defm VSUMWSX : RVF1m<"vsum.w.sx", 0xea, V64, VM>;
let cx2 = 1 in defm VSUMWZX : RVF1m<"vsum.w.zx", 0xea, V64, VM>;
// Section 8.14.2 - VSUMX (Vector Sum)
defm VSUML : RVF1m<"vsum.l", 0xaa, V64, VM>;
// Section 8.14.3 - VFSUM (Vector Floating Sum)
defm VFSUMD : RVF1m<"vfsum.d", 0xec, V64, VM>;
let cx = 1 in defm VFSUMS : RVF1m<"vfsum.s", 0xec, V64, VM>;
// Section 8.14.4 - VMAXS (Vector Maximum/Minimum Single)
let cx2 = 0 in defm VRMAXSWFSTSX : RVF1m<"vrmaxs.w.fst.sx", 0xbb, V64, VM>;
let cx2 = 1 in defm VRMAXSWFSTZX : RVF1m<"vrmaxs.w.fst.zx", 0xbb, V64, VM>;
let cs = 1 in {
let cx2 = 0 in
defm VRMAXSWLSTSX : RVF1m<"vrmaxs.w.lst.sx", 0xbb, V64, VM>;
let cx2 = 1 in
defm VRMAXSWLSTZX : RVF1m<"vrmaxs.w.lst.zx", 0xbb, V64, VM>;
}
let cs2 = 1 in {
let cx2 = 0 in
defm VRMINSWFSTSX : RVF1m<"vrmins.w.fst.sx", 0xbb, V64, VM>;
let cx2 = 1 in
defm VRMINSWFSTZX : RVF1m<"vrmins.w.fst.zx", 0xbb, V64, VM>;
let cs = 1 in {
let cx2 = 0 in
defm VRMINSWLSTSX : RVF1m<"vrmins.w.lst.sx", 0xbb, V64, VM>;
let cx2 = 1 in
defm VRMINSWLSTZX : RVF1m<"vrmins.w.lst.zx", 0xbb, V64, VM>;
}
}
// Section 8.14.5 - VMAXX (Vector Maximum/Minimum)
let cs = 0 in defm VRMAXSLFST : RVF1m<"vrmaxs.l.fst", 0xab, V64, VM>;
let cs = 1 in defm VRMAXSLLST : RVF1m<"vrmaxs.l.lst", 0xab, V64, VM>;
let cs2 = 1 in {
let cs = 0 in defm VRMINSLFST : RVF1m<"vrmins.l.fst", 0xab, V64, VM>;
let cs = 1 in defm VRMINSLLST : RVF1m<"vrmins.l.lst", 0xab, V64, VM>;
}
// Section 8.14.6 - VFMAX (Vector Floating Maximum/Minimum)
let cs = 0 in defm VFRMAXDFST : RVF1m<"vfrmax.d.fst", 0xad, V64, VM>;
let cs = 1 in defm VFRMAXDLST : RVF1m<"vfrmax.d.lst", 0xad, V64, VM>;
let cs2 = 1 in {
let cs = 0 in defm VFRMINDFST : RVF1m<"vfrmin.d.fst", 0xad, V64, VM>;
let cs = 1 in defm VFRMINDLST : RVF1m<"vfrmin.d.lst", 0xad, V64, VM>;
}
let cx = 1 in {
let cs = 0 in defm VFRMAXSFST : RVF1m<"vfrmax.s.fst", 0xad, V64, VM>;
let cs = 1 in defm VFRMAXSLST : RVF1m<"vfrmax.s.lst", 0xad, V64, VM>;
let cs2 = 1 in {
let cs = 0 in defm VFRMINSFST : RVF1m<"vfrmin.s.fst", 0xad, V64, VM>;
let cs = 1 in defm VFRMINSLST : RVF1m<"vfrmin.s.lst", 0xad, V64, VM>;
}
}
// Section 8.14.7 - VRAND (Vector Reduction And)
defm VRAND : RVF1m<"vrand", 0x88, V64, VM>;
// Section 8.14.8 - VROR (Vector Reduction Or)
defm VROR : RVF1m<"vror", 0x98, V64, VM>;
// Section 8.14.9 - VRXOR (Vector Reduction Exclusive Or)
defm VRXOR : RVF1m<"vrxor", 0x89, V64, VM>;

llvm/test/MC/VE/VFMAX.s

  • This file was added.
# RUN: llvm-mc -triple=ve --show-encoding < %s \
# RUN: | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
# RUN: llvm-mc -triple=ve -filetype=obj < %s | llvm-objdump -d - \
# RUN: | FileCheck %s --check-prefixes=CHECK-INST
# CHECK-INST: vfrmax.d.fst %v11, %v12
# CHECK-ENCODING: encoding: [0x00,0x00,0x0c,0x0b,0x00,0x00,0x00,0xad]
vfrmax.d.fst %v11, %v12
# CHECK-INST: vfrmax.d.fst %v11, %vix, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x0b,0xad]
vfrmax.d.fst %v11, %vix, %vm11
# CHECK-INST: vfrmax.d.lst %vix, %v22, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0xff,0x00,0x00,0x2f,0xad]
vfrmax.d.lst %vix, %v22, %vm15
# CHECK-INST: vfrmax.s.lst %v63, %v60, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0x3c,0x3f,0x00,0x00,0xa2,0xad]
vfrmax.s.lst %v63, %v60, %vm2
# CHECK-INST: vfrmax.s.fst %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x80,0xad]
vfrmax.s.fst %vix, %vix, %vm0
# CHECK-INST: vfrmax.s.lst %vix, %vix, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0xa2,0xad]
vfrmax.s.lst %vix, %vix, %vm2
# CHECK-INST: vfrmin.d.fst %v11, %v12
# CHECK-ENCODING: encoding: [0x00,0x00,0x0c,0x0b,0x00,0x00,0x10,0xad]
vfrmin.d.fst %v11, %v12
# CHECK-INST: vfrmin.d.fst %v11, %vix, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x1b,0xad]
vfrmin.d.fst %v11, %vix, %vm11
# CHECK-INST: vfrmin.d.lst %vix, %v22, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0xff,0x00,0x00,0x3f,0xad]
vfrmin.d.lst %vix, %v22, %vm15
# CHECK-INST: vfrmin.s.lst %v63, %v60, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0x3c,0x3f,0x00,0x00,0xb2,0xad]
vfrmin.s.lst %v63, %v60, %vm2
# CHECK-INST: vfrmin.s.fst %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x90,0xad]
vfrmin.s.fst %vix, %vix, %vm0
# CHECK-INST: vfrmin.s.lst %vix, %vix, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0xb2,0xad]
vfrmin.s.lst %vix, %vix, %vm2

llvm/test/MC/VE/VFSUM.s

  • This file was added.
# RUN: llvm-mc -triple=ve --show-encoding < %s \
# RUN: | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
# RUN: llvm-mc -triple=ve -filetype=obj < %s | llvm-objdump -d - \
# RUN: | FileCheck %s --check-prefixes=CHECK-INST
# CHECK-INST: vfsum.d %v11, %v12
# CHECK-ENCODING: encoding: [0x00,0x00,0x0c,0x0b,0x00,0x00,0x00,0xec]
vfsum.d %v11, %v12
# CHECK-INST: vfsum.d %v11, %vix, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x0b,0xec]
vfsum.d %v11, %vix, %vm11
# CHECK-INST: vfsum.d %vix, %v22, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0xff,0x00,0x00,0x0f,0xec]
vfsum.d %vix, %v22, %vm15
# CHECK-INST: vfsum.s %v63, %v60, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0x3c,0x3f,0x00,0x00,0x82,0xec]
vfsum.s %v63, %v60, %vm2
# CHECK-INST: vfsum.s %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x80,0xec]
vfsum.s %vix, %vix, %vm0
# CHECK-INST: vfsum.s %vix, %vix, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x82,0xec]
vfsum.s %vix, %vix, %vm2

llvm/test/MC/VE/VMAXS.s

  • This file was added.
# RUN: llvm-mc -triple=ve --show-encoding < %s \
# RUN: | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
# RUN: llvm-mc -triple=ve -filetype=obj < %s | llvm-objdump -d - \
# RUN: | FileCheck %s --check-prefixes=CHECK-INST
# CHECK-INST: vrmaxs.w.fst.sx %v11, %v12
# CHECK-ENCODING: encoding: [0x00,0x00,0x0c,0x0b,0x00,0x00,0x00,0xbb]
vrmaxs.w.fst.sx %v11, %v12
# CHECK-INST: vrmaxs.w.fst.sx %v11, %vix, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x0b,0xbb]
vrmaxs.w.fst.sx %v11, %vix, %vm11
# CHECK-INST: vrmaxs.w.lst.sx %vix, %v22, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0xff,0x00,0x00,0x2f,0xbb]
vrmaxs.w.lst.sx %vix, %v22, %vm15
# CHECK-INST: vrmaxs.w.lst.zx %v63, %v60, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0x3c,0x3f,0x00,0x00,0x62,0xbb]
vrmaxs.w.lst.zx %v63, %v60, %vm2
# CHECK-INST: vrmaxs.w.fst.zx %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x40,0xbb]
vrmaxs.w.fst.zx %vix, %vix, %vm0
# CHECK-INST: vrmaxs.w.lst.zx %vix, %vix, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x62,0xbb]
vrmaxs.w.lst.zx %vix, %vix, %vm2
# CHECK-INST: vrmins.w.fst.sx %v11, %v12
# CHECK-ENCODING: encoding: [0x00,0x00,0x0c,0x0b,0x00,0x00,0x10,0xbb]
vrmins.w.fst.sx %v11, %v12
# CHECK-INST: vrmins.w.fst.sx %v11, %vix, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x1b,0xbb]
vrmins.w.fst.sx %v11, %vix, %vm11
# CHECK-INST: vrmins.w.lst.sx %vix, %v22, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0xff,0x00,0x00,0x3f,0xbb]
vrmins.w.lst.sx %vix, %v22, %vm15
# CHECK-INST: vrmins.w.lst.zx %v63, %v60, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0x3c,0x3f,0x00,0x00,0x72,0xbb]
vrmins.w.lst.zx %v63, %v60, %vm2
# CHECK-INST: vrmins.w.fst.zx %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x50,0xbb]
vrmins.w.fst.zx %vix, %vix, %vm0
# CHECK-INST: vrmins.w.lst.zx %vix, %vix, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x72,0xbb]
vrmins.w.lst.zx %vix, %vix, %vm2

llvm/test/MC/VE/VRAND.s

  • This file was added.
# RUN: llvm-mc -triple=ve --show-encoding < %s \
# RUN: | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
# RUN: llvm-mc -triple=ve -filetype=obj < %s | llvm-objdump -d - \
# RUN: | FileCheck %s --check-prefixes=CHECK-INST
# CHECK-INST: vrand %v11, %v22
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0x0b,0x00,0x00,0x00,0x88]
vrand %v11, %v22
# CHECK-INST: vrand %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x00,0x88]
vrand %vix, %vix
# CHECK-INST: vrand %v11, %v22, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0x0b,0x00,0x00,0x0b,0x88]
vrand %v11, %v22, %vm11
# CHECK-INST: vrand %v11, %vix, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x0f,0x88]
vrand %v11, %vix, %vm15

llvm/test/MC/VE/VROR.s

  • This file was added.
# RUN: llvm-mc -triple=ve --show-encoding < %s \
# RUN: | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
# RUN: llvm-mc -triple=ve -filetype=obj < %s | llvm-objdump -d - \
# RUN: | FileCheck %s --check-prefixes=CHECK-INST
# CHECK-INST: vror %v11, %v22
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0x0b,0x00,0x00,0x00,0x98]
vror %v11, %v22
# CHECK-INST: vror %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x00,0x98]
vror %vix, %vix
# CHECK-INST: vror %v11, %v22, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0x0b,0x00,0x00,0x0b,0x98]
vror %v11, %v22, %vm11
# CHECK-INST: vror %v11, %vix, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x0f,0x98]
vror %v11, %vix, %vm15

llvm/test/MC/VE/VRXOR.s

  • This file was added.
# RUN: llvm-mc -triple=ve --show-encoding < %s \
# RUN: | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
# RUN: llvm-mc -triple=ve -filetype=obj < %s | llvm-objdump -d - \
# RUN: | FileCheck %s --check-prefixes=CHECK-INST
# CHECK-INST: vrxor %v11, %v22
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0x0b,0x00,0x00,0x00,0x89]
vrxor %v11, %v22
# CHECK-INST: vrxor %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x00,0x89]
vrxor %vix, %vix
# CHECK-INST: vrxor %v11, %v22, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0x0b,0x00,0x00,0x0b,0x89]
vrxor %v11, %v22, %vm11
# CHECK-INST: vrxor %v11, %vix, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x0f,0x89]
vrxor %v11, %vix, %vm15

llvm/test/MC/VE/VSUMS.s

  • This file was added.
# RUN: llvm-mc -triple=ve --show-encoding < %s \
# RUN: | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
# RUN: llvm-mc -triple=ve -filetype=obj < %s | llvm-objdump -d - \
# RUN: | FileCheck %s --check-prefixes=CHECK-INST
# CHECK-INST: vsum.w.sx %v11, %v12
# CHECK-ENCODING: encoding: [0x00,0x00,0x0c,0x0b,0x00,0x00,0x00,0xea]
vsum.w.sx %v11, %v12
# CHECK-INST: vsum.w.sx %v11, %vix, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x0b,0xea]
vsum.w.sx %v11, %vix, %vm11
# CHECK-INST: vsum.w.sx %vix, %v22, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0xff,0x00,0x00,0x0f,0xea]
vsum.w.sx %vix, %v22, %vm15
# CHECK-INST: vsum.w.zx %v63, %v60, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0x3c,0x3f,0x00,0x00,0x42,0xea]
vsum.w.zx %v63, %v60, %vm2
# CHECK-INST: vsum.w.zx %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x40,0xea]
vsum.w.zx %vix, %vix, %vm0
# CHECK-INST: vsum.w.zx %vix, %vix, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x42,0xea]
vsum.w.zx %vix, %vix, %vm2

llvm/test/MC/VE/VSUMX.s

  • This file was added.
# RUN: llvm-mc -triple=ve --show-encoding < %s \
# RUN: | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
# RUN: llvm-mc -triple=ve -filetype=obj < %s | llvm-objdump -d - \
# RUN: | FileCheck %s --check-prefixes=CHECK-INST
# CHECK-INST: vsum.l %v11, %v12
# CHECK-ENCODING: encoding: [0x00,0x00,0x0c,0x0b,0x00,0x00,0x00,0xaa]
vsum.l %v11, %v12
# CHECK-INST: vsum.l %v11, %vix, %vm11
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0x0b,0x00,0x00,0x0b,0xaa]
vsum.l %v11, %vix, %vm11
# CHECK-INST: vsum.l %vix, %v22, %vm15
# CHECK-ENCODING: encoding: [0x00,0x00,0x16,0xff,0x00,0x00,0x0f,0xaa]
vsum.l %vix, %v22, %vm15
# CHECK-INST: vsum.l %v63, %v60, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0x3c,0x3f,0x00,0x00,0x02,0xaa]
vsum.l %v63, %v60, %vm2
# CHECK-INST: vsum.l %vix, %vix
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x00,0xaa]
vsum.l %vix, %vix, %vm0
# CHECK-INST: vsum.l %vix, %vix, %vm2
# CHECK-ENCODING: encoding: [0x00,0x00,0xff,0xff,0x00,0x00,0x02,0xaa]
vsum.l %vix, %vix, %vm2