Skip to content

Commit

Permalink
[RISCV][VLOPT] Add support for 11.11 div instructions (llvm#112201)
Browse files Browse the repository at this point in the history 
This adds support for these instructions and also tests getOperandInfo
for these instructions as well.
  • Loading branch information
@michaelmaitland
michaelmaitland authored Oct 14, 2024
1 parent 82e89c0 commit c2c4db8
Show file tree
Hide file tree
Showing 2 changed files with 168 additions and 1 deletion.
9 changes: 8 additions & 1 deletion llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -554,7 +554,14 @@ static bool isSupportedInstr(const MachineInstr &MI) {
case RISCV::VMULHSU_VV:
case RISCV::VMULHSU_VX:
// 11.11. Vector Integer Divide Instructions
// FIXME: Add support for 11.11 instructions
case RISCV::VDIVU_VV:
case RISCV::VDIVU_VX:
case RISCV::VDIV_VV:
case RISCV::VDIV_VX:
case RISCV::VREMU_VV:
case RISCV::VREMU_VX:
case RISCV::VREM_VV:
case RISCV::VREM_VX:
// 11.12. Vector Widening Integer Multiply Instructions
// FIXME: Add support for 11.12 instructions
// 11.13. Vector Single-Width Integer Multiply-Add Instructions
Expand Down
160 changes: 160 additions & 0 deletions llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -962,6 +962,166 @@ define <vscale x 4 x i32> @vmulhsu_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vdivu_vv(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
; NOVLOPT-LABEL: vdivu_vv:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vdivu.vv v8, v8, v10
; NOVLOPT-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; NOVLOPT-NEXT: vmul.vv v8, v8, v10
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vdivu_vv:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; VLOPT-NEXT: vdivu.vv v8, v8, v10
; VLOPT-NEXT: vmul.vv v8, v8, v10
; VLOPT-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vdivu.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1)
%2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %b, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vdivu_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
; NOVLOPT-LABEL: vdivu_vx:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vdivu.vx v10, v8, a0
; NOVLOPT-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; NOVLOPT-NEXT: vmul.vv v8, v10, v8
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vdivu_vx:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; VLOPT-NEXT: vdivu.vx v10, v8, a0
; VLOPT-NEXT: vmul.vv v8, v10, v8
; VLOPT-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vdivu.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, i32 %b, iXLen -1)
%2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %a, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vdiv_vv(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
; NOVLOPT-LABEL: vdiv_vv:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vdiv.vv v8, v8, v10
; NOVLOPT-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; NOVLOPT-NEXT: vmul.vv v8, v8, v10
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vdiv_vv:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; VLOPT-NEXT: vdiv.vv v8, v8, v10
; VLOPT-NEXT: vmul.vv v8, v8, v10
; VLOPT-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vdiv.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1)
%2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %b, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vdiv_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
; NOVLOPT-LABEL: vdiv_vx:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vdiv.vx v10, v8, a0
; NOVLOPT-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; NOVLOPT-NEXT: vmul.vv v8, v10, v8
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vdiv_vx:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; VLOPT-NEXT: vdiv.vx v10, v8, a0
; VLOPT-NEXT: vmul.vv v8, v10, v8
; VLOPT-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vdiv.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, i32 %b, iXLen -1)
%2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %a, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vremu_vv(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
; NOVLOPT-LABEL: vremu_vv:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vremu.vv v8, v8, v10
; NOVLOPT-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; NOVLOPT-NEXT: vmul.vv v8, v8, v10
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vremu_vv:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; VLOPT-NEXT: vremu.vv v8, v8, v10
; VLOPT-NEXT: vmul.vv v8, v8, v10
; VLOPT-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vremu.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1)
%2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %b, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vremu_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
; NOVLOPT-LABEL: vremu_vx:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vremu.vx v10, v8, a0
; NOVLOPT-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; NOVLOPT-NEXT: vmul.vv v8, v10, v8
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vremu_vx:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; VLOPT-NEXT: vremu.vx v10, v8, a0
; VLOPT-NEXT: vmul.vv v8, v10, v8
; VLOPT-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vremu.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, i32 %b, iXLen -1)
%2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %a, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vrem_vv(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
; NOVLOPT-LABEL: vrem_vv:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vrem.vv v8, v8, v10
; NOVLOPT-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; NOVLOPT-NEXT: vmul.vv v8, v8, v10
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vrem_vv:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; VLOPT-NEXT: vrem.vv v8, v8, v10
; VLOPT-NEXT: vmul.vv v8, v8, v10
; VLOPT-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vrem.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1)
%2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %b, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vrem_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
; NOVLOPT-LABEL: vrem_vx:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vrem.vx v10, v8, a0
; NOVLOPT-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; NOVLOPT-NEXT: vmul.vv v8, v10, v8
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vrem_vx:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; VLOPT-NEXT: vrem.vx v10, v8, a0
; VLOPT-NEXT: vmul.vv v8, v10, v8
; VLOPT-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vrem.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, i32 %b, iXLen -1)
%2 = call <vscale x 4 x i32> @llvm.riscv.vmul.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %a, iXLen %vl)
ret <vscale x 4 x i32> %2
}

define <vscale x 4 x i32> @vwmacc_vx(<vscale x 4 x i16> %a, i16 %b, iXLen %vl) {
; NOVLOPT-LABEL: vwmacc_vx:
; NOVLOPT: # %bb.0:
Expand Down

0 comments on commit c2c4db8

Please sign in to comment.