[RISCV] Add support for V extension in SiFive7

[michaelmaitland]

Summary

Add scheduling information for vector extension in SiFive7,
while using new LMUL & SEW scheduling constructs.

[michaelmaitland]

Remove zfh from x280 check in clang/test/Driver/riscv-cpus.c

[craig.topper]

llvm/lib/Target/RISCV/RISCVProcessors.td

169 ↗

(On Diff #518086)

I would prefer that we add sifive-x280 in a single patch with all of its features. Currently zba/zbb are in a different patch. I don't think there's any restriction about adding a feature to a CPU without it being present in the scheduler model.

[craig.topper]

All CPU name additions should be mentioned in the RISC-V section of llvm/docs/ReleaseNotes.rst

[michaelmaitland]

Split adding sifive-x280 and vector model between this patch and https://reviews.llvm.org/D149710

[craig.topper]

llvm/lib/Target/RISCV/RISCVSchedSiFive7.td

112

There's an extra space before VLUpperBound here

[• pcwang-thead]

llvm/lib/Target/RISCV/RISCVSchedSiFive7.td

15

I think I have fixed the issue that defar can't refer to template arguments in D148197. So LMUL, SSEW and other fields can be replaced with defvars.

llvm/lib/Target/RISCV/RISCVScheduleV.td

41

So, are we going to discard LMULXXXImpl below?

[reames]

LGTM, thought please wait for other review feedback to settle.

I am very very happy to have this upstream, thank you!

[michaelmaitland]

Remove extra space before VLUpperBound.

llvm/lib/Target/RISCV/RISCVSchedSiFive7.td

15

A defvar statement defines a global variable. SiFive7IsWorstCaseMX<mx, SchedMxList>.c becomes invalid syntax since c is not a member of SiFive7IsWorstCaseMX now that c is global. If we did use defvar here and gave c a more unique name so that it could fit in the global space, then we would also need to complicate it by adding a MXxxx_MxListxxx suffix for all mx and SchedMxList pairs.

I think we'd like to keep these as Type Iden = Value;. What do you think?

[• pcwang-thead]

LGTM.
Though I don't like the way that we need to loop MxList again when defining scheduling model, I think this can be a good practice to define RVV scheduling model in current TableGen's grammar.

llvm/lib/Target/RISCV/RISCVSchedSiFive7.td

15

I mean:

class SiFive7IsWorstCaseMX<string mx, list<string> MxList> {
  defvar LLMUL = LargestLMUL<MxList>.r;
  bit c = !eq(mx, LLMUL);
}

LLMUL can be a local variable so that there is only one field c(which is the result of this class-based subroutine). We can keep only one field as result and change other fields to local variables. In this way, we know that SiFive7IsWorstCaseMX is a subroutine conventionally and can be replace with function in the future(if my patches is merged).

[michaelmaitland]

Use defvar in subroutines.

[michaelmaitland]

llvm/lib/Target/RISCV/RISCVScheduleV.td

41

I will use LMULXXXImpl in a follow up NFC patch. This code below comes from before LMULXXXImpl was introduced.

[michaelmaitland]

llvm/lib/Target/RISCV/RISCVScheduleV.td

41

Can you please provide an example of how you intend on this being used today? I think it depends on https://reviews.llvm.org/D146198 since if we use LMULXXXImpl today, there is no way to differentiate behavior based on LMUL.

[• pcwang-thead]

llvm/lib/Target/RISCV/RISCVScheduleV.td

41

An example is the old version of this patch(https://reviews.llvm.org/D146198?vs=on&id=509607#toc) and the comment(https://reviews.llvm.org/D146198?vs=on&id=509607#4232814).
Maybe you can wait for the approval of these patches if we're gonna change it anyway. So I think there is no need to change it currently. :-)

[reames]

Let's go ahead and land this as is, we can rework the stylistic pieces once the linked patches land.

[michaelmaitland]

In D149495#4373768, @reames wrote:

Let's go ahead and land this as is, we can rework the stylistic pieces once the linked patches land.

This has landed in https://reviews.llvm.org/rG1a855819a87f426bdbd83c815fa47ca01fdf928f. I raised my question regarding examples in my previous comment on this post because I am starting to rework the stylistic concerns raised by @pcwang-thead. However, I will wait until Wang's patches land before making any modifications to style.

[reames]

In D149495#4373796, @michaelmaitland wrote:

In D149495#4373768, @reames wrote:

Let's go ahead and land this as is, we can rework the stylistic pieces once the linked patches land.

This has landed in https://reviews.llvm.org/rG1a855819a87f426bdbd83c815fa47ca01fdf928f. I raised my question regarding examples in my previous comment on this post because I am starting to rework the stylistic concerns raised by @pcwang-thead. However, I will wait until Wang's patches land before making any modifications to style.

My apologies, I read right past the commit in the history. Oops. Sorry for the noise!

llvm/lib/Target/RISCV/RISCVSchedSiFive7.td

 //==- RISCVSchedSiFive7.td - SiFive7 Scheduling Definitions --*- tablegen -*-=//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
 
+/// c is true if mx has the worst case behavior compared to LMULs in MxList.
+/// On the SiFive7, the worst case LMUL is the Largest LMUL
+/// and the worst case sew is the smallest SEW for that LMUL.
+class SiFive7IsWorstCaseMX<string mx, list<string> MxList> {
+  defvar LLMUL = LargestLMUL<MxList>.r;
+  bit c = !eq(mx, LLMUL);
+}
+
+/// c is true if mx and sew have the worst case behavior compared to LMULs in
+/// MxList. On the SiFive7, the worst case LMUL is the Largest LMUL
+/// and the worst case sew is the smallest SEW for that LMUL.
+class SiFive7IsWorstCaseMXSEW<string mx, int sew, list<string> MxList,
+                               bit isF = 0> {
+  defvar LLMUL = LargestLMUL<MxList>.r;
+  defvar SSEW = SmallestSEW<mx, isF>.r;
+  bit c = !and(!eq(mx, LLMUL), !eq(sew, SSEW));
+}
+
+class SiFive7GetCyclesDefault<string mx> {
+  int c = !cond(
+    !eq(mx, "M1") : 2,
+    !eq(mx, "M2") : 4,
+    !eq(mx, "M4") : 8,
+    !eq(mx, "M8") : 16,
+    !eq(mx, "MF2") : 1,
+    !eq(mx, "MF4") : 1,
+    !eq(mx, "MF8") : 1
+  );
+}
+
+class SiFive7GetCyclesWidening<string mx> {
+  int c = !cond(
+    !eq(mx, "M1") : 2,
+    !eq(mx, "M2") : 4,
+    !eq(mx, "M4") : 8,
+    !eq(mx, "MF2") : 1,
+    !eq(mx, "MF4") : 1,
+    !eq(mx, "MF8") : 1
+  );
+}
+
+class SiFive7GetCyclesNarrowing<string mx> {
+  int c = !cond(
+    !eq(mx, "M1") : 4,
+    !eq(mx, "M2") : 8,
+    !eq(mx, "M4") : 16,
+    !eq(mx, "MF2") : 2,
+    !eq(mx, "MF4") : 1,
+    !eq(mx, "MF8") : 1
+  );
+}
+
+class SiFive7GetCyclesOutputLMUL<string mx> {
+  int c = !cond(
+    !eq(mx, "M1") : 1,
+    !eq(mx, "M2") : 2,
+    !eq(mx, "M4") : 4,
+    !eq(mx, "M8") : 8,
+    !eq(mx, "MF2") : 1,
+    !eq(mx, "MF4") : 1,
+    !eq(mx, "MF8") : 1
+  );
+}
+
+class SiFive7GetCyclesVMask<string mx> {
+  int c = !cond(
+    !eq(mx, "M1") : 1,
+    !eq(mx, "M2") : 1,
+    !eq(mx, "M4") : 1,
+    !eq(mx, "M8") : 2,
+    !eq(mx, "MF2") : 1,
+    !eq(mx, "MF4") : 1,
+    !eq(mx, "MF8") : 1
+  );
+}
+
+// Cycles for segmented loads and stores are calculated using the
+// formula ceil(2 * nf * lmul).
+class SiFive7GetCyclesSegmented<string mx, int nf> {
+  int c = !cond(
+    !eq(mx, "M1") : !mul(!mul(2, nf), 1),
+    !eq(mx, "M2") : !mul(!mul(2, nf), 2),
+    !eq(mx, "M4") : !mul(!mul(2, nf), 4),
+    !eq(mx, "M8") : !mul(!mul(2, nf), 8),
+    // We can calculate ceil(a/b) using (a + b - 1) / b.
+    // Since the multiplication of fractional lmul is the
+    // same as division by the denominator the formula we
+    // use is ceil(2 * nf / lmul_denominator). We can use
+    // ceil(a/b) where a = 2 * nf, b = lmul_denominator.
+    !eq(mx, "MF2") : !div(!sub(!add(!mul(2, nf), 2), 1), 2),
+    !eq(mx, "MF4") : !div(!sub(!add(!mul(2, nf), 4), 1), 4),
+    !eq(mx, "MF8") : !div(!sub(!add(!mul(2, nf), 8), 1), 8)
+  );
+}
+
+class SiFive7GetCyclesOnePerElement<string mx, int sew> {
+  // FIXME: On SiFive7, VLEN is 512. Although a user can request the compiler
+  // to use a different VLEN, this model will not make scheduling decisions
+  // based on the user specified VLEN.
+  // c = ceil(VLEN / SEW) * LMUL
+  // Note: c >= 1 since the smallest VLUpperBound is 512 / 8 = 8, and the
+  // largest division performed on VLUpperBound is in MF8 case with division
+  // by 8. Therefore, there is no need to ceil the result.
+  int VLUpperBound = !div(512, sew);
+  int c = !cond(
+    !eq(mx, "M1") :        VLUpperBound,
+    !eq(mx, "M2") :        !mul(VLUpperBound, 2),
+    !eq(mx, "M4") :        !mul(VLUpperBound, 4),
+    !eq(mx, "M8") :        !mul(VLUpperBound, 8),
+    !eq(mx, "MF2") :       !div(VLUpperBound, 2),
+    !eq(mx, "MF4") :       !div(VLUpperBound, 4),
+    !eq(mx, "MF8") :       !div(VLUpperBound, 8)
+  );
+}
+
+class SiFive7GetDivOrSqrtFactor<int sew> {
+  int c = !cond(
+    // TODO: Add SchedSEWSetFP upstream and remove the SEW=8 case.
+    !eq(sew, 8) : 15,
+    !eq(sew, 16) : 15,
+    !eq(sew, 32) : 28,
+    !eq(sew, 64) : 57
+  );
+}
+
 // SiFive7 machine model for scheduling and other instruction cost heuristics.
 def SiFive7Model : SchedMachineModel {
   let MicroOpBufferSize = 0; // Explicitly set to zero since SiFive7 is in-order.
   let IssueWidth = 2;        // 2 micro-ops are dispatched per cycle.
   let LoadLatency = 3;
   let MispredictPenalty = 3;
   let CompleteModel = 0;
   let UnsupportedFeatures = [HasStdExtZbkb, HasStdExtZbkc, HasStdExtZbkx,
                              HasStdExtZcmt, HasStdExtZknd, HasStdExtZkne,
                              HasStdExtZknh, HasStdExtZksed, HasStdExtZksh,
-                             HasStdExtZkr, HasVInstructions];
+                             HasStdExtZkr];
 }
 
-// The SiFive7 microarchitecture has two pipelines: A and B.
+// The SiFive7 microarchitecture has three pipelines: A, B, V.
 // Pipe A can handle memory, integer alu and vector operations.
 // Pipe B can handle integer alu, control flow, integer multiply and divide,
 // and floating point computation.
+// Pipe V can handle the V extension.
 let SchedModel = SiFive7Model in {
 let BufferSize = 0 in {
 def SiFive7PipeA       : ProcResource<1>;
 def SiFive7PipeB       : ProcResource<1>;
+def SiFive7PipeV       : ProcResource<1>;
 }
 
 let BufferSize = 1 in {
 def SiFive7IDiv        : ProcResource<1> { let Super = SiFive7PipeB; } // Int Division
 def SiFive7FDiv        : ProcResource<1> { let Super = SiFive7PipeB; } // FP Division/Sqrt
+def SiFive7VA          : ProcResource<1> { let Super = SiFive7PipeV; } // Arithmetic sequencer
+def SiFive7VL          : ProcResource<1> { let Super = SiFive7PipeV; } // Load sequencer
+def SiFive7VS          : ProcResource<1> { let Super = SiFive7PipeV; } // Store sequencer
 }
 
 def SiFive7PipeAB : ProcResGroup<[SiFive7PipeA, SiFive7PipeB]>;
 
 // Branching
 def : WriteRes<WriteJmp, [SiFive7PipeB]>;
 def : WriteRes<WriteJal, [SiFive7PipeB]>;
 def : WriteRes<WriteJalr, [SiFive7PipeB]>;

 def : WriteRes<WriteFMovI16ToF16, [SiFive7PipeB]>;
 def : WriteRes<WriteFMovF16ToI16, [SiFive7PipeB]>;
 def : WriteRes<WriteFMovI32ToF32, [SiFive7PipeB]>;
 def : WriteRes<WriteFMovF32ToI32, [SiFive7PipeB]>;
 def : WriteRes<WriteFMovI64ToF64, [SiFive7PipeB]>;
 def : WriteRes<WriteFMovF64ToI64, [SiFive7PipeB]>;
 }
 
+// 6. Configuration-Setting Instructions
+let Latency = 3 in {
+def : WriteRes<WriteVSETVLI, [SiFive7PipeA]>;
+def : WriteRes<WriteVSETIVLI, [SiFive7PipeA]>;
+def : WriteRes<WriteVSETVL, [SiFive7PipeA]>;
+}
+
+// 7. Vector Loads and Stores
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesDefault<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = Cycles, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVLDE",    [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTE",    [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDM",    [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTM",    [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDS8",   [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDS16",  [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDS32",  [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDS64",  [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTS8",   [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTS16",  [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTS32",  [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTS64",  [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDUX8",  [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDUX16", [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDUX32", [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDUX64", [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDOX8",  [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDOX16", [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDOX32", [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDOX64", [SiFive7VL], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTUX8",  [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTUX16", [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTUX32", [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTUX64", [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTOX8",  [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTOX16", [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTOX32", [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSTOX64", [SiFive7VS], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVLDFF",   [SiFive7VL], mx, IsWorstCase>;
+  }
+}
+
+// VLD*R is LMUL aware
+let Latency = 4, ResourceCycles = [2] in
+  def : WriteRes<WriteVLD1R,  [SiFive7VL]>;
+let Latency = 4, ResourceCycles = [4] in
+  def : WriteRes<WriteVLD2R,  [SiFive7VL]>;
+let Latency = 4, ResourceCycles = [8] in
+  def : WriteRes<WriteVLD4R,  [SiFive7VL]>;
+let Latency = 4, ResourceCycles = [16] in
+  def : WriteRes<WriteVLD8R,  [SiFive7VL]>;
+// VST*R is LMUL aware
+let Latency = 1, ResourceCycles = [2] in
+  def : WriteRes<WriteVST1R,   [SiFive7VS]>;
+let Latency = 1, ResourceCycles = [4] in
+  def : WriteRes<WriteVST2R,   [SiFive7VS]>;
+let Latency = 1, ResourceCycles = [8] in
+  def : WriteRes<WriteVST4R,   [SiFive7VS]>;
+let Latency = 1, ResourceCycles = [16] in
+  def : WriteRes<WriteVST8R,   [SiFive7VS]>;
+
+foreach mx = SchedMxList in {
+  foreach nf=2-8 in {
+    foreach eew = [8, 16, 32, 64] in {
+      defvar Cycles = SiFive7GetCyclesSegmented<mx, nf>.c;
+      defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+      let Latency = Cycles, ResourceCycles = [Cycles] in {
+        defm "" : LMULWriteResMX<"WriteVLSEG" # nf # "e" # eew,   [SiFive7VL], mx, IsWorstCase>;
+        defm "" : LMULWriteResMX<"WriteVLSEGFF" # nf # "e" # eew, [SiFive7VL], mx, IsWorstCase>;
+        defm "" : LMULWriteResMX<"WriteVLSSEG" # nf # "e" # eew,  [SiFive7VL], mx, IsWorstCase>;
+        defm "" : LMULWriteResMX<"WriteVLUXSEG" # nf # "e" # eew, [SiFive7VL], mx, IsWorstCase>;
+        defm "" : LMULWriteResMX<"WriteVLOXSEG" # nf # "e" # eew, [SiFive7VL], mx, IsWorstCase>;
+        defm "" : LMULWriteResMX<"WriteVSSEG" # nf # "e" # eew,   [SiFive7VS], mx, IsWorstCase>;
+        defm "" : LMULWriteResMX<"WriteVSSSEG" # nf # "e" # eew,  [SiFive7VS], mx, IsWorstCase>;
+        defm "" : LMULWriteResMX<"WriteVSUXSEG" # nf # "e" # eew, [SiFive7VS], mx, IsWorstCase>;
+        defm "" : LMULWriteResMX<"WriteVSOXSEG" # nf # "e" # eew, [SiFive7VS], mx, IsWorstCase>;
+      }
+    }
+  }
+}
+
+// 11. Vector Integer Arithmetic Instructions
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesDefault<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = 4, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVIALUV",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIALUX",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIALUI",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVICALUV",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVICALUX",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVICALUI",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVShiftV",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVShiftX",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVShiftI",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVICmpV",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVICmpX",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVICmpI",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMulV",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMulX",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMulAddV",  [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMulAddX",  [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMergeV",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMergeX",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMergeI",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMovV",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMovX",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMovI",     [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesOutputLMUL<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = 4, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVExtV",      [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+foreach mx = SchedMxList in {
+  foreach sew = SchedSEWSet<mx>.val in {
+    defvar Cycles = !mul(SiFive7GetDivOrSqrtFactor<sew>.c,
+                         !div(SiFive7GetCyclesOnePerElement<mx, sew>.c, 4));
+    defvar IsWorstCase = SiFive7IsWorstCaseMXSEW<mx, sew, SchedMxList>.c;
+    let Latency = Cycles, ResourceCycles = [Cycles] in {
+      defm "" : LMULSEWWriteResMXSEW<"WriteVIDivV", [SiFive7VA], mx, sew, IsWorstCase>;
+      defm "" : LMULSEWWriteResMXSEW<"WriteVIDivX", [SiFive7VA], mx, sew, IsWorstCase>;
+    }
+  }
+}
+
+// Widening
+foreach mx = SchedMxListW in {
+  defvar Cycles = SiFive7GetCyclesWidening<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxListW>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVIWALUV",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIWALUX",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIWALUI",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIWMulV",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIWMulX",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIWMulAddV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIWMulAddX", [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+// Narrowing
+foreach mx = SchedMxListW in {
+  defvar Cycles = SiFive7GetCyclesNarrowing<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxListW>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVNShiftV",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVNShiftX",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVNShiftI",   [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+
+// 12. Vector Fixed-Point Arithmetic Instructions
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesDefault<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVSALUV",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSALUX",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSALUI",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVAALUV",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVAALUX",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSMulV",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSMulX",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSShiftV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSShiftX", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVSShiftI", [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+// Narrowing
+foreach mx = SchedMxListW in {
+  defvar Cycles = SiFive7GetCyclesNarrowing<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxListW>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVNClipV",  [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVNClipX",  [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVNClipI",  [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+
+// 13. Vector Floating-Point Instructions
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesDefault<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVFALUV",      [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFALUF",      [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFMulV",      [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFMulF",      [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFMulAddV",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFMulAddF",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFRecpV",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFCmpV",      [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFCmpF",      [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFSgnjV",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFSgnjF",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFClassV",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFMergeV",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFMovV",      [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFCvtIToFV",  [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFCvtFToIV",  [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+foreach mx = SchedMxListF in {
+  foreach sew = SchedSEWSetF<mx>.val in {
+    defvar Cycles = !mul(SiFive7GetDivOrSqrtFactor<sew>.c,
+                         !div(SiFive7GetCyclesOnePerElement<mx, sew>.c, 4));
+    defvar IsWorstCase = SiFive7IsWorstCaseMXSEW<mx, sew, SchedMxListF, 1>.c;
+    let Latency = Cycles, ResourceCycles = [Cycles] in {
+      defm "" : LMULSEWWriteResMXSEW<"WriteVFSqrtV", [SiFive7VA], mx, sew, IsWorstCase>;
+      defm "" : LMULSEWWriteResMXSEW<"WriteVFDivV",  [SiFive7VA], mx, sew, IsWorstCase>;
+      defm "" : LMULSEWWriteResMXSEW<"WriteVFDivF",  [SiFive7VA], mx, sew, IsWorstCase>;
+    }
+  }
+}
+
+// Widening
+foreach mx = SchedMxListW in {
+  defvar Cycles = SiFive7GetCyclesWidening<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxListW>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVFWCvtIToFV", [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+foreach mx = SchedMxListFW in {
+  defvar Cycles = SiFive7GetCyclesWidening<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxListFW>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVFWALUV",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFWMulV",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFWMulAddV",  [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFWCvtFToIV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFWCvtFToFV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFWMulAddF",  [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFWMulF",     [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFWALUF",     [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+// Narrowing
+foreach mx = SchedMxListW in {
+  defvar Cycles = SiFive7GetCyclesNarrowing<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxListW>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVFNCvtFToIV", [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+foreach mx = SchedMxListFW in {
+  defvar Cycles = SiFive7GetCyclesNarrowing<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxListFW>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVFNCvtIToFV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFNCvtFToFV", [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+
+// 14. Vector Reduction Operations
+let Latency = 32 in {
+defm "" : LMULWriteRes<"WriteVIRedV_From", [SiFive7VA]>;
+defm "" : LMULWriteRes<"WriteVIWRedV_From", [SiFive7VA]>;
+defm "" : LMULWriteRes<"WriteVFRedV_From", [SiFive7VA]>;
+defm "" : LMULWriteRes<"WriteVFRedOV_From", [SiFive7VA]>;
+defm "" : LMULWriteResFWRed<"WriteVFWRedV_From", [SiFive7VA]>;
+defm "" : LMULWriteResFWRed<"WriteVFWRedOV_From", [SiFive7VA]>;
+}
+
+// 15. Vector Mask Instructions
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesVMask<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = 4, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVMALUV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVMPopV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVMFFSV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVMSFSV", [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesDefault<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = 4, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVMIotV", [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVMIdxV", [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+
+// 16. Vector Permutation Instructions
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesDefault<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = 8, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVIMovVX",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVIMovXV",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFMovVF",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFMovFV",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVRGatherVX",    [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVRGatherVI",    [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+
+foreach mx = SchedMxList in {
+  foreach sew = SchedSEWSet<mx>.val in {
+    defvar Cycles = SiFive7GetCyclesOnePerElement<mx, sew>.c;
+    defvar IsWorstCase = SiFive7IsWorstCaseMXSEW<mx, sew, SchedMxList>.c;
+    let Latency = !add(Cycles, 3), ResourceCycles = [Cycles] in {
+      defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherVV", [SiFive7VA], mx, sew, IsWorstCase>;
+      defm "" : LMULSEWWriteResMXSEW<"WriteVCompressV", [SiFive7VA], mx, sew, IsWorstCase>;
+    }
+  }
+}
+
+foreach mx = SchedMxList in {
+  defvar Cycles = SiFive7GetCyclesDefault<mx>.c;
+  defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
+  let Latency = 4, ResourceCycles = [Cycles] in {
+    defm "" : LMULWriteResMX<"WriteVISlideX",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVISlideI",   [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVISlide1X",  [SiFive7VA], mx, IsWorstCase>;
+    defm "" : LMULWriteResMX<"WriteVFSlide1F",  [SiFive7VA], mx, IsWorstCase>;
+  }
+}
+
+// VMov*V is LMUL Aware
+let Latency = 4, ResourceCycles = [2] in
+  def : WriteRes<WriteVMov1V,     [SiFive7VA]>;
+let Latency = 4, ResourceCycles = [4] in
+  def : WriteRes<WriteVMov2V,     [SiFive7VA]>;
+let Latency = 4, ResourceCycles = [8] in
+  def : WriteRes<WriteVMov4V,     [SiFive7VA]>;
+let Latency = 4, ResourceCycles = [16] in
+  def : WriteRes<WriteVMov8V,     [SiFive7VA]>;
+
 // Others
 def : WriteRes<WriteCSR, [SiFive7PipeB]>;
 def : WriteRes<WriteNop, []>;
+let Latency = 3 in
+  def : WriteRes<WriteRdVLENB, [SiFive7PipeB]>;
 
 def : InstRW<[WriteIALU], (instrs COPY)>;
 
 //===----------------------------------------------------------------------===//
 // Bypass and advance
 def : ReadAdvance<ReadJmp, 0>;
 def : ReadAdvance<ReadJalr, 0>;
 def : ReadAdvance<ReadCSR, 0>;

 def : ReadAdvance<ReadCTZ32, 0>;
 def : ReadAdvance<ReadCPOP, 0>;
 def : ReadAdvance<ReadCPOP32, 0>;
 def : ReadAdvance<ReadORCB, 0>;
 def : ReadAdvance<ReadREV8, 0>;
 def : ReadAdvance<ReadSHXADD, 0>;
 def : ReadAdvance<ReadSHXADD32, 0>;
 
+// 6. Configuration-Setting Instructions
+def : ReadAdvance<ReadVSETVLI, 2>;
+def : ReadAdvance<ReadVSETVL, 2>;
+
+// 7. Vector Loads and Stores
+def : ReadAdvance<ReadVLDX, 0>;
+def : ReadAdvance<ReadVSTX, 0>;
+defm "" : LMULReadAdvance<"ReadVSTEV", 0>;
+defm "" : LMULReadAdvance<"ReadVSTM", 0>;
+def : ReadAdvance<ReadVLDSX, 0>;
+def : ReadAdvance<ReadVSTSX, 0>;
+defm "" : LMULReadAdvance<"ReadVSTS8V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTS16V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTS32V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTS64V", 0>;
+defm "" : LMULReadAdvance<"ReadVLDUXV", 0>;
+defm "" : LMULReadAdvance<"ReadVLDOXV", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUX8", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUX16", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUX32", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUX64", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUXV", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUX8V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUX16V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUX32V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTUX64V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOX8", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOX16", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOX32", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOX64", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOXV", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOX8V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOX16V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOX32V", 0>;
+defm "" : LMULReadAdvance<"ReadVSTOX64V", 0>;
+// LMUL Aware
+def : ReadAdvance<ReadVST1R, 0>;
+def : ReadAdvance<ReadVST2R, 0>;
+def : ReadAdvance<ReadVST4R, 0>;
+def : ReadAdvance<ReadVST8R, 0>;
+
+// 12. Vector Integer Arithmetic Instructions
+defm : LMULReadAdvance<"ReadVIALUV", 0>;
+defm : LMULReadAdvance<"ReadVIALUX", 0>;
+defm : LMULReadAdvanceW<"ReadVIWALUV", 0>;
+defm : LMULReadAdvanceW<"ReadVIWALUX", 0>;
+defm : LMULReadAdvance<"ReadVExtV", 0>;
+defm : LMULReadAdvance<"ReadVICALUV", 0>;
+defm : LMULReadAdvance<"ReadVICALUX", 0>;
+defm : LMULReadAdvance<"ReadVShiftV", 0>;
+defm : LMULReadAdvance<"ReadVShiftX", 0>;
+defm : LMULReadAdvanceW<"ReadVNShiftV", 0>;
+defm : LMULReadAdvanceW<"ReadVNShiftX", 0>;
+defm : LMULReadAdvance<"ReadVICmpV", 0>;
+defm : LMULReadAdvance<"ReadVICmpX", 0>;
+defm : LMULReadAdvance<"ReadVIMulV", 0>;
+defm : LMULReadAdvance<"ReadVIMulX", 0>;
+defm : LMULSEWReadAdvance<"ReadVIDivV", 0>;
+defm : LMULSEWReadAdvance<"ReadVIDivX", 0>;
+defm : LMULReadAdvanceW<"ReadVIWMulV", 0>;
+defm : LMULReadAdvanceW<"ReadVIWMulX", 0>;
+defm : LMULReadAdvance<"ReadVIMulAddV", 0>;
+defm : LMULReadAdvance<"ReadVIMulAddX", 0>;
+defm : LMULReadAdvanceW<"ReadVIWMulAddV", 0>;
+defm : LMULReadAdvanceW<"ReadVIWMulAddX", 0>;
+defm : LMULReadAdvance<"ReadVIMergeV", 0>;
+defm : LMULReadAdvance<"ReadVIMergeX", 0>;
+defm : LMULReadAdvance<"ReadVIMovV", 0>;
+defm : LMULReadAdvance<"ReadVIMovX", 0>;
+
+// 13. Vector Fixed-Point Arithmetic Instructions
+defm "" : LMULReadAdvance<"ReadVSALUV", 0>;
+defm "" : LMULReadAdvance<"ReadVSALUX", 0>;
+defm "" : LMULReadAdvance<"ReadVAALUV", 0>;
+defm "" : LMULReadAdvance<"ReadVAALUX", 0>;
+defm "" : LMULReadAdvance<"ReadVSMulV", 0>;
+defm "" : LMULReadAdvance<"ReadVSMulX", 0>;
+defm "" : LMULReadAdvance<"ReadVSShiftV", 0>;
+defm "" : LMULReadAdvance<"ReadVSShiftX", 0>;
+defm "" : LMULReadAdvanceW<"ReadVNClipV", 0>;
+defm "" : LMULReadAdvanceW<"ReadVNClipX", 0>;
+
+// 14. Vector Floating-Point Instructions
+defm "" : LMULReadAdvance<"ReadVFALUV", 0>;
+defm "" : LMULReadAdvance<"ReadVFALUF", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFWALUV", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFWALUF", 0>;
+defm "" : LMULReadAdvance<"ReadVFMulV", 0>;
+defm "" : LMULReadAdvance<"ReadVFMulF", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFDivV", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFDivF", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFWMulV", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFWMulF", 0>;
+defm "" : LMULReadAdvance<"ReadVFMulAddV", 0>;
+defm "" : LMULReadAdvance<"ReadVFMulAddF", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFWMulAddV", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFWMulAddF", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFSqrtV", 0>;
+defm "" : LMULReadAdvance<"ReadVFRecpV", 0>;
+defm "" : LMULReadAdvance<"ReadVFCmpV", 0>;
+defm "" : LMULReadAdvance<"ReadVFCmpF", 0>;
+defm "" : LMULReadAdvance<"ReadVFSgnjV", 0>;
+defm "" : LMULReadAdvance<"ReadVFSgnjF", 0>;
+defm "" : LMULReadAdvance<"ReadVFClassV", 0>;
+defm "" : LMULReadAdvance<"ReadVFMergeV", 0>;
+defm "" : LMULReadAdvance<"ReadVFMergeF", 0>;
+defm "" : LMULReadAdvance<"ReadVFMovF", 0>;
+defm "" : LMULReadAdvance<"ReadVFCvtIToFV", 0>;
+defm "" : LMULReadAdvance<"ReadVFCvtFToIV", 0>;
+defm "" : LMULReadAdvanceW<"ReadVFWCvtIToFV", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFWCvtFToIV", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFWCvtFToFV", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFNCvtIToFV", 0>;
+defm "" : LMULReadAdvanceW<"ReadVFNCvtFToIV", 0>;
+defm "" : LMULReadAdvanceFW<"ReadVFNCvtFToFV", 0>;
+
+// 15. Vector Reduction Operations
+def : ReadAdvance<ReadVIRedV, 0>;
+def : ReadAdvance<ReadVIRedV0, 0>;
+def : ReadAdvance<ReadVIWRedV, 0>;
+def : ReadAdvance<ReadVIWRedV0, 0>;
+def : ReadAdvance<ReadVFRedV, 0>;
+def : ReadAdvance<ReadVFRedV0, 0>;
+def : ReadAdvance<ReadVFRedOV, 0>;
+def : ReadAdvance<ReadVFRedOV0, 0>;
+def : ReadAdvance<ReadVFWRedV, 0>;
+def : ReadAdvance<ReadVFWRedV0, 0>;
+def : ReadAdvance<ReadVFWRedOV, 0>;
+def : ReadAdvance<ReadVFWRedOV0, 0>;
+
+// 16. Vector Mask Instructions
+defm "" : LMULReadAdvance<"ReadVMALUV", 0>;
+defm "" : LMULReadAdvance<"ReadVMPopV", 0>;
+defm "" : LMULReadAdvance<"ReadVMFFSV", 0>;
+defm "" : LMULReadAdvance<"ReadVMSFSV", 0>;
+defm "" : LMULReadAdvance<"ReadVMIotV", 0>;
+
+// 17. Vector Permutation Instructions
+defm "" : LMULReadAdvance<"ReadVIMovVX", 0>;
+defm "" : LMULReadAdvance<"ReadVIMovXV", 0>;
+defm "" : LMULReadAdvance<"ReadVIMovXX", 0>;
+defm "" : LMULReadAdvance<"ReadVFMovVF", 0>;
+defm "" : LMULReadAdvance<"ReadVFMovFV", 0>;
+defm "" : LMULReadAdvance<"ReadVFMovFX", 0>;
+defm "" : LMULReadAdvance<"ReadVISlideV", 0>;
+defm "" : LMULReadAdvance<"ReadVISlideX", 0>;
+defm "" : LMULReadAdvance<"ReadVFSlideV", 0>;
+defm "" : LMULReadAdvance<"ReadVFSlideF", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_data", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_index", 0>;
+defm "" : LMULReadAdvance<"ReadVRGatherVX_data", 0>;
+defm "" : LMULReadAdvance<"ReadVRGatherVX_index", 0>;
+defm "" : LMULReadAdvance<"ReadVRGatherVI_data", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVCompressV", 0>;
+// LMUL Aware
+def : ReadAdvance<ReadVMov1V, 0>;
+def : ReadAdvance<ReadVMov2V, 0>;
+def : ReadAdvance<ReadVMov4V, 0>;
+def : ReadAdvance<ReadVMov8V, 0>;
+
+// Others
+def : ReadAdvance<ReadVMask, 0>;
+
 //===----------------------------------------------------------------------===//
 // Unsupported extensions
-defm : UnsupportedSchedV;
 defm : UnsupportedSchedZbc;
 defm : UnsupportedSchedZbs;
 defm : UnsupportedSchedZbkb;
 defm : UnsupportedSchedZbkx;
 defm : UnsupportedSchedZfa;
 }

llvm/lib/Target/RISCV/RISCVScheduleV.td

 //===- RISCVScheduleV.td - RISC-V Scheduling Definitions V -*- tablegen -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
 /// Define scheduler resources associated with def operands.
 
-defvar SchedMxList = ["M1", "M2", "M4", "M8", "MF2", "MF4", "MF8"];
+defvar SchedMxList = ["MF8", "MF4", "MF2", "M1", "M2", "M4", "M8"];
 // Used for widening and narrowing instructions as it doesn't contain M8.
 defvar SchedMxListW = !listremove(SchedMxList, ["M8"]);
 defvar SchedMxListFW = !listremove(SchedMxList, ["M8", "MF8"]);
 // Used for floating-point as it doesn't contain MF8.
 defvar SchedMxListF = !listremove(SchedMxList, ["MF8"]);
 // Used for widening floating-point Reduction as it doesn't contain MF8.
 defvar SchedMxListFWRed = SchedMxListF;
 
class SchedSEWSetF<string mx> {
   list<int> val = !cond(!eq(mx, "M1"):  [16, 32, 64],
                         !eq(mx, "M2"):  [16, 32, 64],
                         !eq(mx, "M4"):  [16, 32, 64],
                         !eq(mx, "M8"):  [16, 32, 64],
                         !eq(mx, "MF2"): [16, 32],
                         !eq(mx, "MF4"): [16]);
 }
 
+// Helper function to get the largest LMUL from MxList
+// Precondition: MxList is sorted in ascending LMUL order.
+class LargestLMUL<list<string> MxList> {
+  // MX list is sorted from smallest to largest
+  string r = !foldl(!head(MxList), MxList, last, curr, curr);
+}
+// Helper function to get the smallest SEW that can be used with LMUL mx
+// Precondition: MxList is sorted in ascending LMUL order and SchedSEWSet<mx>
+class SmallestSEW<string mx, bit isF = 0> {
+  int r = !head(!if(isF, SchedSEWSetF<mx>.val, SchedSEWSet<mx>.val));
+}
+
+// Creates WriteRes for (name, mx, resources) tuple
+multiclass LMULWriteResMX<string name, list<ProcResourceKind> resources,
+                          string mx, bit IsWorstCase> {
+  def : WriteRes<!cast<SchedWrite>(name # "_" # mx), resources>;
+  if IsWorstCase then
+    def : WriteRes<!cast<SchedWrite>(name # "_WorstCase"), resources>;
+}
+multiclass LMULSEWWriteResMXSEW<string name, list<ProcResourceKind> resources,
+                             string mx, int sew,  bit IsWorstCase> {
+  def : WriteRes<!cast<SchedWrite>(name # "_" # mx # "_E" # sew), resources>;
+  if IsWorstCase then
+    def : WriteRes<!cast<SchedWrite>(name # "_WorstCase"), resources>;
+}
+
 // Define multiclasses to define SchedWrite, SchedRead,  WriteRes, and
 // ReadAdvance for each (name, LMUL) pair and for each LMUL in each of the
 // SchedMxList variants above. Each multiclass is responsible for defining
 // a record that represents the WorseCase behavior for name.
 multiclass LMULSchedWritesImpl<string name, list<string> MxList> {
   def name # "_WorstCase" : SchedWrite;
   foreach mx = MxList in {
     def name # "_" # mx : SchedWrite;