[RISCV][VLOPT] support fp sign injection instructions

[michaelmaitland]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

Author: Michael Maitland (michaelmaitland)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/124195.diff

2 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp (+7)
• (modified) llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll (+120)

diff --git a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
index 5f2d4e0585a0b3..3031721c6988af 100644
--- a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
+++ b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
@@ -1017,6 +1017,13 @@ static bool isSupportedInstr(const MachineInstr &MI) {
   // Vector Widening Floating-Point Multiply
   case RISCV::VFWMUL_VF:
   case RISCV::VFWMUL_VV:
+  // Vector Floating-Point Sign-Injection Instructions
+  case RISCV::VFSGNJ_VF:
+  case RISCV::VFSGNJ_VV:
+  case RISCV::VFSGNJN_VV:
+  case RISCV::VFSGNJN_VF:
+  case RISCV::VFSGNJX_VF:
+  case RISCV::VFSGNJX_VV:
   // Vector Floating-Point Compare Instructions
   case RISCV::VMFEQ_VF:
   case RISCV::VMFEQ_VV:
diff --git a/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll b/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
index ce94e1c193645a..6ebf0c717d1efa 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
@@ -3831,3 +3831,123 @@ define <vscale x 4 x i32> @vasubu_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
   %2 = call <vscale x 4 x i32> @llvm.riscv.vadd.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 float> @vfsgnj_vv(<vscale x 4 x float> %a, <vscale x 4 x float> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vfsgnj_vv:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfsgnj.vv v8, v8, v10
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfadd.vv v8, v8, v10
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vfsgnj_vv:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vfsgnj.vv v8, v8, v10
+; VLOPT-NEXT:    vfadd.vv v8, v8, v10
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x float> @llvm.riscv.vfsgnj.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %a, <vscale x 4 x float> %b, iXLen -1)
+  %2 = call <vscale x 4 x float> @llvm.riscv.vfadd.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %1, <vscale x 4 x float> %b, iXLen 7, iXLen %vl)
+  ret <vscale x 4 x float> %2
+}
+
+define <vscale x 4 x float> @vfsgnj_vf(<vscale x 4 x float> %a, float %b, iXLen %vl) {
+; NOVLOPT-LABEL: vfsgnj_vf:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfsgnj.vf v10, v8, fa0
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfadd.vv v8, v10, v8
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vfsgnj_vf:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vfsgnj.vf v10, v8, fa0
+; VLOPT-NEXT:    vfadd.vv v8, v10, v8
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x float> @llvm.riscv.vfsgnj.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %a, float %b, iXLen -1)
+  %2 = call <vscale x 4 x float> @llvm.riscv.vfadd.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %1, <vscale x 4 x float> %a, iXLen 7, iXLen %vl)
+  ret <vscale x 4 x float> %2
+}
+
+define <vscale x 4 x float> @vfsgnjn_vv(<vscale x 4 x float> %a, <vscale x 4 x float> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vfsgnjn_vv:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfsgnjn.vv v8, v8, v10
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfadd.vv v8, v8, v10
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vfsgnjn_vv:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vfsgnjn.vv v8, v8, v10
+; VLOPT-NEXT:    vfadd.vv v8, v8, v10
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x float> @llvm.riscv.vfsgnjn.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %a, <vscale x 4 x float> %b, iXLen -1)
+  %2 = call <vscale x 4 x float> @llvm.riscv.vfadd.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %1, <vscale x 4 x float> %b, iXLen 7, iXLen %vl)
+  ret <vscale x 4 x float> %2
+}
+
+define <vscale x 4 x float> @vfsgnjn_vf(<vscale x 4 x float> %a, float %b, iXLen %vl) {
+; NOVLOPT-LABEL: vfsgnjn_vf:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfsgnjn.vf v10, v8, fa0
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfadd.vv v8, v10, v8
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vfsgnjn_vf:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vfsgnjn.vf v10, v8, fa0
+; VLOPT-NEXT:    vfadd.vv v8, v10, v8
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x float> @llvm.riscv.vfsgnjn.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %a, float %b, iXLen -1)
+  %2 = call <vscale x 4 x float> @llvm.riscv.vfadd.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %1, <vscale x 4 x float> %a, iXLen 7, iXLen %vl)
+  ret <vscale x 4 x float> %2
+}
+
+define <vscale x 4 x float> @vfsgnjx_vv(<vscale x 4 x float> %a, <vscale x 4 x float> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vfsgnjx_vv:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfsgnjx.vv v8, v8, v10
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfadd.vv v8, v8, v10
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vfsgnjx_vv:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vfsgnjx.vv v8, v8, v10
+; VLOPT-NEXT:    vfadd.vv v8, v8, v10
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x float> @llvm.riscv.vfsgnjx.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %a, <vscale x 4 x float> %b, iXLen -1)
+  %2 = call <vscale x 4 x float> @llvm.riscv.vfadd.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %1, <vscale x 4 x float> %b, iXLen 7, iXLen %vl)
+  ret <vscale x 4 x float> %2
+}
+
+define <vscale x 4 x float> @vfsgnjx_vf(<vscale x 4 x float> %a, float %b, iXLen %vl) {
+; NOVLOPT-LABEL: vfsgnjx_vf:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfsgnjx.vf v10, v8, fa0
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vfadd.vv v8, v10, v8
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vfsgnjx_vf:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vfsgnjx.vf v10, v8, fa0
+; VLOPT-NEXT:    vfadd.vv v8, v10, v8
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x float> @llvm.riscv.vfsgnjx.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %a, float %b, iXLen -1)
+  %2 = call <vscale x 4 x float> @llvm.riscv.vfadd.nxv4f32.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> %1, <vscale x 4 x float> %a, iXLen 7, iXLen %vl)
+  ret <vscale x 4 x float> %2
+}

[preames]

LGTM