mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-22 18:54:02 +01:00
[RISCV] Lower CONCAT_VECTORS to INSERT_SUBVECTOR nodes
The default expansion of CONCAT_VECTORS goes through the stack. This patch avoids that penalty by custom-lowering CONCAT_VECTORS to a series of INSERT_SUBVECTOR nodes. Futher optimizations are possible, but this is a good start. Reviewed By: craig.topper Differential Revision: https://reviews.llvm.org/D97692
This commit is contained in:
parent
3c9c0a88f1
commit
d1351b0b35
@ -536,6 +536,7 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
|
||||
setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Custom);
|
||||
|
||||
setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
|
||||
setOperationAction(ISD::CONCAT_VECTORS, VT, Custom);
|
||||
|
||||
setOperationAction(ISD::LOAD, VT, Custom);
|
||||
setOperationAction(ISD::STORE, VT, Custom);
|
||||
@ -1494,6 +1495,20 @@ SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
|
||||
return lowerBUILD_VECTOR(Op, DAG, Subtarget);
|
||||
case ISD::VECTOR_SHUFFLE:
|
||||
return lowerVECTOR_SHUFFLE(Op, DAG, Subtarget);
|
||||
case ISD::CONCAT_VECTORS: {
|
||||
// Split CONCAT_VECTORS into a series of INSERT_SUBVECTOR nodes. This is
|
||||
// better than going through the stack, as the default expansion does.
|
||||
SDLoc DL(Op);
|
||||
MVT VT = Op.getSimpleValueType();
|
||||
assert(VT.isFixedLengthVector() && "Unexpected CONCAT_VECTORS lowering");
|
||||
unsigned NumOpElts =
|
||||
Op.getOperand(0).getSimpleValueType().getVectorNumElements();
|
||||
SDValue Vec = DAG.getUNDEF(VT);
|
||||
for (const auto &OpIdx : enumerate(Op->ops()))
|
||||
Vec = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, VT, Vec, OpIdx.value(),
|
||||
DAG.getIntPtrConstant(OpIdx.index() * NumOpElts, DL));
|
||||
return Vec;
|
||||
}
|
||||
case ISD::LOAD:
|
||||
return lowerFixedLengthVectorLoadToRVV(Op, DAG);
|
||||
case ISD::STORE:
|
||||
|
@ -295,15 +295,13 @@ define void @fp2si_v8f64_v8i8(<8 x double>* %x, <8 x i8>* %y) {
|
||||
;
|
||||
; LMULMAX1-LABEL: fp2si_v8f64_v8i8:
|
||||
; LMULMAX1: # %bb.0:
|
||||
; LMULMAX1-NEXT: addi sp, sp, -16
|
||||
; LMULMAX1-NEXT: .cfi_def_cfa_offset 16
|
||||
; LMULMAX1-NEXT: vsetivli a2, 2, e64,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vle64.v v25, (a0)
|
||||
; LMULMAX1-NEXT: addi a2, a0, 16
|
||||
; LMULMAX1-NEXT: vsetivli a3, 2, e64,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vle64.v v25, (a2)
|
||||
; LMULMAX1-NEXT: vle64.v v26, (a0)
|
||||
; LMULMAX1-NEXT: addi a2, a0, 32
|
||||
; LMULMAX1-NEXT: vle64.v v26, (a2)
|
||||
; LMULMAX1-NEXT: addi a2, a0, 48
|
||||
; LMULMAX1-NEXT: vle64.v v27, (a2)
|
||||
; LMULMAX1-NEXT: addi a0, a0, 16
|
||||
; LMULMAX1-NEXT: addi a0, a0, 48
|
||||
; LMULMAX1-NEXT: vle64.v v28, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.x.f.w v29, v27
|
||||
@ -311,49 +309,43 @@ define void @fp2si_v8f64_v8i8(<8 x double>* %x, <8 x i8>* %y) {
|
||||
; LMULMAX1-NEXT: vnsrl.wi v27, v29, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v29, v27, 0
|
||||
; LMULMAX1-NEXT: addi a0, sp, 6
|
||||
; LMULMAX1-NEXT: vsetivli a2, 2, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vse8.v v29, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.x.f.w v27, v28
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v28, v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v27, v28, 0
|
||||
; LMULMAX1-NEXT: addi a0, sp, 2
|
||||
; LMULMAX1-NEXT: vsetivli a2, 2, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vse8.v v27, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.x.f.w v27, v26
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v26, v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v27, v26, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a0, sp, 4
|
||||
; LMULMAX1-NEXT: vse8.v v27, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a0, sp, 4
|
||||
; LMULMAX1-NEXT: vle8.v v26, (a0)
|
||||
; LMULMAX1-NEXT: addi a0, sp, 12
|
||||
; LMULMAX1-NEXT: vse8.v v26, (a0)
|
||||
; LMULMAX1-NEXT: vmv.v.i v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vmv1r.v v30, v27
|
||||
; LMULMAX1-NEXT: vslideup.vi v30, v29, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.x.f.w v29, v28
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v28, v29, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v29, v28, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v30, v29, 2
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.x.f.w v28, v26
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v26, v28, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v28, v26, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v27, v28, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.x.f.w v26, v25
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v25, v26, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v26, v25, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vse8.v v26, (sp)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vle8.v v25, (sp)
|
||||
; LMULMAX1-NEXT: addi a0, sp, 8
|
||||
; LMULMAX1-NEXT: vse8.v v25, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v27, v26, 2
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vmv.v.i v25, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v25, v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v25, v30, 4
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a0, sp, 8
|
||||
; LMULMAX1-NEXT: vle8.v v25, (a0)
|
||||
; LMULMAX1-NEXT: vse8.v v25, (a1)
|
||||
; LMULMAX1-NEXT: addi sp, sp, 16
|
||||
; LMULMAX1-NEXT: ret
|
||||
%a = load <8 x double>, <8 x double>* %x
|
||||
%d = fptosi <8 x double> %a to <8 x i8>
|
||||
@ -378,15 +370,13 @@ define void @fp2ui_v8f64_v8i8(<8 x double>* %x, <8 x i8>* %y) {
|
||||
;
|
||||
; LMULMAX1-LABEL: fp2ui_v8f64_v8i8:
|
||||
; LMULMAX1: # %bb.0:
|
||||
; LMULMAX1-NEXT: addi sp, sp, -16
|
||||
; LMULMAX1-NEXT: .cfi_def_cfa_offset 16
|
||||
; LMULMAX1-NEXT: vsetivli a2, 2, e64,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vle64.v v25, (a0)
|
||||
; LMULMAX1-NEXT: addi a2, a0, 16
|
||||
; LMULMAX1-NEXT: vsetivli a3, 2, e64,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vle64.v v25, (a2)
|
||||
; LMULMAX1-NEXT: vle64.v v26, (a0)
|
||||
; LMULMAX1-NEXT: addi a2, a0, 32
|
||||
; LMULMAX1-NEXT: vle64.v v26, (a2)
|
||||
; LMULMAX1-NEXT: addi a2, a0, 48
|
||||
; LMULMAX1-NEXT: vle64.v v27, (a2)
|
||||
; LMULMAX1-NEXT: addi a0, a0, 16
|
||||
; LMULMAX1-NEXT: addi a0, a0, 48
|
||||
; LMULMAX1-NEXT: vle64.v v28, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.xu.f.w v29, v27
|
||||
@ -394,49 +384,43 @@ define void @fp2ui_v8f64_v8i8(<8 x double>* %x, <8 x i8>* %y) {
|
||||
; LMULMAX1-NEXT: vnsrl.wi v27, v29, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v29, v27, 0
|
||||
; LMULMAX1-NEXT: addi a0, sp, 6
|
||||
; LMULMAX1-NEXT: vsetivli a2, 2, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vse8.v v29, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.xu.f.w v27, v28
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v28, v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v27, v28, 0
|
||||
; LMULMAX1-NEXT: addi a0, sp, 2
|
||||
; LMULMAX1-NEXT: vsetivli a2, 2, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vse8.v v27, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.xu.f.w v27, v26
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v26, v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v27, v26, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a0, sp, 4
|
||||
; LMULMAX1-NEXT: vse8.v v27, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a0, sp, 4
|
||||
; LMULMAX1-NEXT: vle8.v v26, (a0)
|
||||
; LMULMAX1-NEXT: addi a0, sp, 12
|
||||
; LMULMAX1-NEXT: vse8.v v26, (a0)
|
||||
; LMULMAX1-NEXT: vmv.v.i v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vmv1r.v v30, v27
|
||||
; LMULMAX1-NEXT: vslideup.vi v30, v29, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.xu.f.w v29, v28
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v28, v29, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v29, v28, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v30, v29, 2
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.xu.f.w v28, v26
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v26, v28, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v28, v26, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v27, v28, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e32,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vfncvt.rtz.xu.f.w v26, v25
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v25, v26, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v26, v25, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 2, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vse8.v v26, (sp)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vle8.v v25, (sp)
|
||||
; LMULMAX1-NEXT: addi a0, sp, 8
|
||||
; LMULMAX1-NEXT: vse8.v v25, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v27, v26, 2
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vmv.v.i v25, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v25, v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v25, v30, 4
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a0, sp, 8
|
||||
; LMULMAX1-NEXT: vle8.v v25, (a0)
|
||||
; LMULMAX1-NEXT: vse8.v v25, (a1)
|
||||
; LMULMAX1-NEXT: addi sp, sp, 16
|
||||
; LMULMAX1-NEXT: ret
|
||||
%a = load <8 x double>, <8 x double>* %x
|
||||
%d = fptoui <8 x double> %a to <8 x i8>
|
||||
|
@ -211,31 +211,26 @@ define void @trunc_v8i8_v8i32(<8 x i32>* %x, <8 x i8>* %z) {
|
||||
;
|
||||
; LMULMAX1-LABEL: trunc_v8i8_v8i32:
|
||||
; LMULMAX1: # %bb.0:
|
||||
; LMULMAX1-NEXT: addi sp, sp, -16
|
||||
; LMULMAX1-NEXT: .cfi_def_cfa_offset 16
|
||||
; LMULMAX1-NEXT: vsetivli a2, 4, e32,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a2, a0, 16
|
||||
; LMULMAX1-NEXT: vle32.v v25, (a2)
|
||||
; LMULMAX1-NEXT: vle32.v v25, (a0)
|
||||
; LMULMAX1-NEXT: addi a0, a0, 16
|
||||
; LMULMAX1-NEXT: vle32.v v26, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v27, v25, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v25, v27, 0
|
||||
; LMULMAX1-NEXT: addi a0, sp, 12
|
||||
; LMULMAX1-NEXT: vsetivli a2, 4, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vse8.v v25, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: vmv.v.i v27, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v27, v25, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e16,mf2,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v25, v26, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,mf4,ta,mu
|
||||
; LMULMAX1-NEXT: vnsrl.wi v26, v25, 0
|
||||
; LMULMAX1-NEXT: vsetivli a0, 4, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a0, sp, 8
|
||||
; LMULMAX1-NEXT: vse8.v v26, (a0)
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,tu,mu
|
||||
; LMULMAX1-NEXT: vslideup.vi v27, v26, 4
|
||||
; LMULMAX1-NEXT: vsetivli a0, 8, e8,m1,ta,mu
|
||||
; LMULMAX1-NEXT: addi a0, sp, 8
|
||||
; LMULMAX1-NEXT: vle8.v v25, (a0)
|
||||
; LMULMAX1-NEXT: vse8.v v25, (a1)
|
||||
; LMULMAX1-NEXT: addi sp, sp, 16
|
||||
; LMULMAX1-NEXT: vse8.v v27, (a1)
|
||||
; LMULMAX1-NEXT: ret
|
||||
%a = load <8 x i32>, <8 x i32>* %x
|
||||
%b = trunc <8 x i32> %a to <8 x i8>
|
||||
|
Loading…
Reference in New Issue
Block a user