[llvm] 083798e - [LegalizeTypes][VP] Add integer promotion support for vp.fptosi/vp.fptoui
Ben Shi via llvm-commits
llvm-commits at lists.llvm.org
Sun May 29 20:06:01 PDT 2022
Author: Ping Deng
Date: 2022-05-30T03:05:39Z
New Revision: 083798e2702f9b12828ba9d3ee4d1486340a2cb5
URL: https://github.com/llvm/llvm-project/commit/083798e2702f9b12828ba9d3ee4d1486340a2cb5
DIFF: https://github.com/llvm/llvm-project/commit/083798e2702f9b12828ba9d3ee4d1486340a2cb5.diff
LOG: [LegalizeTypes][VP] Add integer promotion support for vp.fptosi/vp.fptoui
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D125760
Added:
Modified:
llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp2i.ll
llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptosi-vp.ll
llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptoui-vp.ll
llvm/test/CodeGen/RISCV/rvv/vfptoi-sdnode.ll
llvm/test/CodeGen/RISCV/rvv/vfptosi-vp.ll
llvm/test/CodeGen/RISCV/rvv/vfptoui-vp.ll
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 686de3933c5f9..123ccfd3550f9 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -135,6 +135,8 @@ void DAGTypeLegalizer::PromoteIntegerResult(SDNode *N, unsigned ResNo) {
case ISD::ZERO_EXTEND:
case ISD::ANY_EXTEND: Res = PromoteIntRes_INT_EXTEND(N); break;
+ case ISD::VP_FPTOSI:
+ case ISD::VP_FPTOUI:
case ISD::STRICT_FP_TO_SINT:
case ISD::STRICT_FP_TO_UINT:
case ISD::FP_TO_SINT:
@@ -669,6 +671,11 @@ SDValue DAGTypeLegalizer::PromoteIntRes_FP_TO_XINT(SDNode *N) {
TLI.isOperationLegalOrCustom(ISD::STRICT_FP_TO_SINT, NVT))
NewOpc = ISD::STRICT_FP_TO_SINT;
+ if (N->getOpcode() == ISD::VP_FPTOUI &&
+ !TLI.isOperationLegal(ISD::VP_FPTOUI, NVT) &&
+ TLI.isOperationLegalOrCustom(ISD::VP_FPTOSI, NVT))
+ NewOpc = ISD::VP_FPTOSI;
+
SDValue Res;
if (N->isStrictFPOpcode()) {
Res = DAG.getNode(NewOpc, dl, {NVT, MVT::Other},
@@ -676,8 +683,12 @@ SDValue DAGTypeLegalizer::PromoteIntRes_FP_TO_XINT(SDNode *N) {
// Legalize the chain result - switch anything that used the old chain to
// use the new one.
ReplaceValueWith(SDValue(N, 1), Res.getValue(1));
- } else
+ } else if (NewOpc == ISD::VP_FPTOSI || NewOpc == ISD::VP_FPTOUI) {
+ Res = DAG.getNode(NewOpc, dl, NVT, {N->getOperand(0), N->getOperand(1),
+ N->getOperand(2)});
+ } else {
Res = DAG.getNode(NewOpc, dl, NVT, N->getOperand(0));
+ }
// Assert that the converted value fits in the original type. If it doesn't
// (eg: because the value being converted is too big), then the result of the
@@ -687,8 +698,11 @@ SDValue DAGTypeLegalizer::PromoteIntRes_FP_TO_XINT(SDNode *N) {
// before legalization: fp-to-uint16, 65534. -> 0xfffe
// after legalization: fp-to-sint32, 65534. -> 0x0000fffe
return DAG.getNode((N->getOpcode() == ISD::FP_TO_UINT ||
- N->getOpcode() == ISD::STRICT_FP_TO_UINT) ?
- ISD::AssertZext : ISD::AssertSext, dl, NVT, Res,
+ N->getOpcode() == ISD::STRICT_FP_TO_UINT ||
+ N->getOpcode() == ISD::VP_FPTOUI)
+ ? ISD::AssertZext
+ : ISD::AssertSext,
+ dl, NVT, Res,
DAG.getValueType(N->getValueType(0).getScalarType()));
}
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp2i.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp2i.ll
index 566fcdbc2ca9a..5a7edd2734ff2 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp2i.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp2i.ll
@@ -44,6 +44,28 @@ define <2 x i1> @fp2si_v2f32_v2i1(<2 x float> %x) {
ret <2 x i1> %z
}
+define <2 x i15> @fp2si_v2f32_v2i15(<2 x float> %x) {
+; CHECK-LABEL: fp2si_v2f32_v2i15:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfncvt.rtz.x.f.w v9, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %z = fptosi <2 x float> %x to <2 x i15>
+ ret <2 x i15> %z
+}
+
+define <2 x i15> @fp2ui_v2f32_v2i15(<2 x float> %x) {
+; CHECK-LABEL: fp2ui_v2f32_v2i15:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfncvt.rtz.x.f.w v9, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %z = fptoui <2 x float> %x to <2 x i15>
+ ret <2 x i15> %z
+}
+
define <2 x i1> @fp2ui_v2f32_v2i1(<2 x float> %x) {
; CHECK-LABEL: fp2ui_v2f32_v2i1:
; CHECK: # %bb.0:
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptosi-vp.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptosi-vp.ll
index d833523079e37..423e123ce30eb 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptosi-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptosi-vp.ll
@@ -4,6 +4,19 @@
; RUN: llc -mtriple=riscv64 -mattr=+m,+v,+zfh,+experimental-zvfh \
; RUN: -riscv-v-vector-bits-min=128 < %s | FileCheck %s
+declare <4 x i7> @llvm.vp.fptosi.v4i7.v4f16(<4 x half>, <4 x i1>, i32)
+
+define <4 x i7> @vfptosi_v4i7_v4f16(<4 x half> %va, <4 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vfptosi_v4i7_v4f16:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT: vfncvt.rtz.x.f.w v9, v8, v0.t
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %v = call <4 x i7> @llvm.vp.fptosi.v4i7.v4f16(<4 x half> %va, <4 x i1> %m, i32 %evl)
+ ret <4 x i7> %v
+}
+
declare <4 x i8> @llvm.vp.fptosi.v4i8.v4f16(<4 x half>, <4 x i1>, i32)
define <4 x i8> @vfptosi_v4i8_v4f16(<4 x half> %va, <4 x i1> %m, i32 zeroext %evl) {
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptoui-vp.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptoui-vp.ll
index 0f96789c6cc93..2f97d38b2a0df 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptoui-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fptoui-vp.ll
@@ -4,6 +4,19 @@
; RUN: llc -mtriple=riscv64 -mattr=+m,+v,+zfh,+experimental-zvfh \
; RUN: -riscv-v-vector-bits-min=128 < %s | FileCheck %s
+declare <4 x i7> @llvm.vp.fptoui.v4i7.v4f16(<4 x half>, <4 x i1>, i32)
+
+define <4 x i7> @vfptoui_v4i7_v4f16(<4 x half> %va, <4 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vfptoui_v4i7_v4f16:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT: vfncvt.rtz.x.f.w v9, v8, v0.t
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %v = call <4 x i7> @llvm.vp.fptoui.v4i7.v4f16(<4 x half> %va, <4 x i1> %m, i32 %evl)
+ ret <4 x i7> %v
+}
+
declare <4 x i8> @llvm.vp.fptoui.v4i8.v4f16(<4 x half>, <4 x i1>, i32)
define <4 x i8> @vfptoui_v4i8_v4f16(<4 x half> %va, <4 x i1> %m, i32 zeroext %evl) {
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfptoi-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfptoi-sdnode.ll
index 680811037399b..d73b6f6f9c9d9 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfptoi-sdnode.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfptoi-sdnode.ll
@@ -16,6 +16,28 @@ define <vscale x 1 x i1> @vfptosi_nxv1f16_nxv1i1(<vscale x 1 x half> %va) {
ret <vscale x 1 x i1> %evec
}
+define <vscale x 1 x i7> @vfptosi_nxv1f16_nxv1i7(<vscale x 1 x half> %va) {
+; CHECK-LABEL: vfptosi_nxv1f16_nxv1i7:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e8, mf8, ta, mu
+; CHECK-NEXT: vfncvt.rtz.x.f.w v9, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %evec = fptosi <vscale x 1 x half> %va to <vscale x 1 x i7>
+ ret <vscale x 1 x i7> %evec
+}
+
+define <vscale x 1 x i7> @vfptoui_nxv1f16_nxv1i7(<vscale x 1 x half> %va) {
+; CHECK-LABEL: vfptoui_nxv1f16_nxv1i7:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e8, mf8, ta, mu
+; CHECK-NEXT: vfncvt.rtz.x.f.w v9, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %evec = fptoui <vscale x 1 x half> %va to <vscale x 1 x i7>
+ ret <vscale x 1 x i7> %evec
+}
+
define <vscale x 1 x i1> @vfptoui_nxv1f16_nxv1i1(<vscale x 1 x half> %va) {
; CHECK-LABEL: vfptoui_nxv1f16_nxv1i1:
; CHECK: # %bb.0:
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfptosi-vp.ll b/llvm/test/CodeGen/RISCV/rvv/vfptosi-vp.ll
index 9686bc1850aff..72d0d400217f3 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfptosi-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfptosi-vp.ll
@@ -2,6 +2,19 @@
; RUN: llc -mtriple=riscv32 -mattr=+m,+v,+zfh,+experimental-zvfh < %s | FileCheck %s
; RUN: llc -mtriple=riscv64 -mattr=+m,+v,+zfh,+experimental-zvfh < %s | FileCheck %s
+declare <vscale x 2 x i7> @llvm.vp.fptosi.v4i7.v4f16(<vscale x 2 x half>, <vscale x 2 x i1>, i32)
+
+define <vscale x 2 x i7> @vfptosi_v4i7_v4f16(<vscale x 2 x half> %va, <vscale x 2 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vfptosi_v4i7_v4f16:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT: vfncvt.rtz.x.f.w v9, v8, v0.t
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %v = call <vscale x 2 x i7> @llvm.vp.fptosi.v4i7.v4f16(<vscale x 2 x half> %va, <vscale x 2 x i1> %m, i32 %evl)
+ ret <vscale x 2 x i7> %v
+}
+
declare <vscale x 2 x i8> @llvm.vp.fptosi.nxv2i8.nxv2f16(<vscale x 2 x half>, <vscale x 2 x i1>, i32)
define <vscale x 2 x i8> @vfptosi_nxv2i8_nxv2f16(<vscale x 2 x half> %va, <vscale x 2 x i1> %m, i32 zeroext %evl) {
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfptoui-vp.ll b/llvm/test/CodeGen/RISCV/rvv/vfptoui-vp.ll
index 9a64cd8d9e3b6..1d1995a6c29a8 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfptoui-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfptoui-vp.ll
@@ -2,6 +2,19 @@
; RUN: llc -mtriple=riscv32 -mattr=+m,+v,+zfh,+experimental-zvfh < %s | FileCheck %s
; RUN: llc -mtriple=riscv64 -mattr=+m,+v,+zfh,+experimental-zvfh < %s | FileCheck %s
+declare <vscale x 2 x i7> @llvm.vp.fptoui.v4i7.v4f16(<vscale x 2 x half>, <vscale x 2 x i1>, i32)
+
+define <vscale x 2 x i7> @vfptoui_v4i7_v4f16(<vscale x 2 x half> %va, <vscale x 2 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vfptoui_v4i7_v4f16:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT: vfncvt.rtz.x.f.w v9, v8, v0.t
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %v = call <vscale x 2 x i7> @llvm.vp.fptoui.v4i7.v4f16(<vscale x 2 x half> %va, <vscale x 2 x i1> %m, i32 %evl)
+ ret <vscale x 2 x i7> %v
+}
+
declare <vscale x 2 x i8> @llvm.vp.fptoui.nxv2i8.nxv2f16(<vscale x 2 x half>, <vscale x 2 x i1>, i32)
define <vscale x 2 x i8> @vfptoui_nxv2i8_nxv2f16(<vscale x 2 x half> %va, <vscale x 2 x i1> %m, i32 zeroext %evl) {
More information about the llvm-commits
mailing list