[RISCV] Add patterns for scalable-vector fsqrt

This patch adds support for lowering the sqrt intrinsic to the RVV
vfsqrt instruction.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D96012

diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
index 3b0d04a..09b33f6 100644 (file)
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
@@ -585,9 +585,14 @@ foreach fvti = AllFloatVectors in {
                  fvti.AVL, fvti.SEW)>;
 }
 
-// 14.10. Vector Floating-Point Sign-Injection Instructions
-// Handle fneg with VFSGNJN using the same input for both operands.
 foreach vti = AllFloatVectors in {
+  // 14.8. Vector Floating-Point Square-Root Instruction
+  def : Pat<(fsqrt (vti.Vector vti.RegClass:$rs2)),
+            (!cast<Instruction>("PseudoVFSQRT_V_"# vti.LMul.MX)
+                 vti.RegClass:$rs2, vti.AVL, vti.SEW)>;
+
+  // 14.10. Vector Floating-Point Sign-Injection Instructions
+  // Handle fneg with VFSGNJN using the same input for both operands.
   def : Pat<(fneg (vti.Vector vti.RegClass:$rs)),
             (!cast<Instruction>("PseudoVFSGNJN_VV_"# vti.LMul.MX)
                  vti.RegClass:$rs, vti.RegClass:$rs, vti.AVL, vti.SEW)>;

diff --git a/llvm/test/CodeGen/RISCV/rvv/vfsqrt-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfsqrt-sdnode.ll
new file mode 100644 (file)
index 0000000..8677967
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/vfsqrt-sdnode.ll
@@ -0,0 +1,185 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=ilp32d \
+; RUN:     -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=lp64d \
+; RUN:     -verify-machineinstrs < %s | FileCheck %s
+
+declare <vscale x 1 x half> @llvm.sqrt.nxv1f16(<vscale x 1 x half>)
+
+define <vscale x 1 x half> @vfsqrt_nxv1f16(<vscale x 1 x half> %v) {
+; CHECK-LABEL: vfsqrt_nxv1f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e16,mf4,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 1 x half> @llvm.sqrt.nxv1f16(<vscale x 1 x half> %v)
+  ret <vscale x 1 x half> %r
+}
+
+declare <vscale x 2 x half> @llvm.sqrt.nxv2f16(<vscale x 2 x half>)
+
+define <vscale x 2 x half> @vfsqrt_nxv2f16(<vscale x 2 x half> %v) {
+; CHECK-LABEL: vfsqrt_nxv2f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e16,mf2,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 2 x half> @llvm.sqrt.nxv2f16(<vscale x 2 x half> %v)
+  ret <vscale x 2 x half> %r
+}
+
+declare <vscale x 4 x half> @llvm.sqrt.nxv4f16(<vscale x 4 x half>)
+
+define <vscale x 4 x half> @vfsqrt_nxv4f16(<vscale x 4 x half> %v) {
+; CHECK-LABEL: vfsqrt_nxv4f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e16,m1,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 4 x half> @llvm.sqrt.nxv4f16(<vscale x 4 x half> %v)
+  ret <vscale x 4 x half> %r
+}
+
+declare <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half>)
+
+define <vscale x 8 x half> @vfsqrt_nxv8f16(<vscale x 8 x half> %v) {
+; CHECK-LABEL: vfsqrt_nxv8f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e16,m2,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half> %v)
+  ret <vscale x 8 x half> %r
+}
+
+declare <vscale x 16 x half> @llvm.sqrt.nxv16f16(<vscale x 16 x half>)
+
+define <vscale x 16 x half> @vfsqrt_nxv16f16(<vscale x 16 x half> %v) {
+; CHECK-LABEL: vfsqrt_nxv16f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e16,m4,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 16 x half> @llvm.sqrt.nxv16f16(<vscale x 16 x half> %v)
+  ret <vscale x 16 x half> %r
+}
+
+declare <vscale x 32 x half> @llvm.sqrt.nxv32f16(<vscale x 32 x half>)
+
+define <vscale x 32 x half> @vfsqrt_nxv32f16(<vscale x 32 x half> %v) {
+; CHECK-LABEL: vfsqrt_nxv32f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e16,m8,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 32 x half> @llvm.sqrt.nxv32f16(<vscale x 32 x half> %v)
+  ret <vscale x 32 x half> %r
+}
+
+declare <vscale x 1 x float> @llvm.sqrt.nxv1f32(<vscale x 1 x float>)
+
+define <vscale x 1 x float> @vfsqrt_nxv1f32(<vscale x 1 x float> %v) {
+; CHECK-LABEL: vfsqrt_nxv1f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e32,mf2,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 1 x float> @llvm.sqrt.nxv1f32(<vscale x 1 x float> %v)
+  ret <vscale x 1 x float> %r
+}
+
+declare <vscale x 2 x float> @llvm.sqrt.nxv2f32(<vscale x 2 x float>)
+
+define <vscale x 2 x float> @vfsqrt_nxv2f32(<vscale x 2 x float> %v) {
+; CHECK-LABEL: vfsqrt_nxv2f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e32,m1,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 2 x float> @llvm.sqrt.nxv2f32(<vscale x 2 x float> %v)
+  ret <vscale x 2 x float> %r
+}
+
+declare <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float>)
+
+define <vscale x 4 x float> @vfsqrt_nxv4f32(<vscale x 4 x float> %v) {
+; CHECK-LABEL: vfsqrt_nxv4f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e32,m2,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float> %v)
+  ret <vscale x 4 x float> %r
+}
+
+declare <vscale x 8 x float> @llvm.sqrt.nxv8f32(<vscale x 8 x float>)
+
+define <vscale x 8 x float> @vfsqrt_nxv8f32(<vscale x 8 x float> %v) {
+; CHECK-LABEL: vfsqrt_nxv8f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e32,m4,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 8 x float> @llvm.sqrt.nxv8f32(<vscale x 8 x float> %v)
+  ret <vscale x 8 x float> %r
+}
+
+declare <vscale x 16 x float> @llvm.sqrt.nxv16f32(<vscale x 16 x float>)
+
+define <vscale x 16 x float> @vfsqrt_nxv16f32(<vscale x 16 x float> %v) {
+; CHECK-LABEL: vfsqrt_nxv16f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e32,m8,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 16 x float> @llvm.sqrt.nxv16f32(<vscale x 16 x float> %v)
+  ret <vscale x 16 x float> %r
+}
+
+declare <vscale x 1 x double> @llvm.sqrt.nxv1f64(<vscale x 1 x double>)
+
+define <vscale x 1 x double> @vfsqrt_nxv1f64(<vscale x 1 x double> %v) {
+; CHECK-LABEL: vfsqrt_nxv1f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e64,m1,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 1 x double> @llvm.sqrt.nxv1f64(<vscale x 1 x double> %v)
+  ret <vscale x 1 x double> %r
+}
+
+declare <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double>)
+
+define <vscale x 2 x double> @vfsqrt_nxv2f64(<vscale x 2 x double> %v) {
+; CHECK-LABEL: vfsqrt_nxv2f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e64,m2,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double> %v)
+  ret <vscale x 2 x double> %r
+}
+
+declare <vscale x 4 x double> @llvm.sqrt.nxv4f64(<vscale x 4 x double>)
+
+define <vscale x 4 x double> @vfsqrt_nxv4f64(<vscale x 4 x double> %v) {
+; CHECK-LABEL: vfsqrt_nxv4f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e64,m4,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 4 x double> @llvm.sqrt.nxv4f64(<vscale x 4 x double> %v)
+  ret <vscale x 4 x double> %r
+}
+
+declare <vscale x 8 x double> @llvm.sqrt.nxv8f64(<vscale x 8 x double>)
+
+define <vscale x 8 x double> @vfsqrt_nxv8f64(<vscale x 8 x double> %v) {
+; CHECK-LABEL: vfsqrt_nxv8f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a0, zero, e64,m8,ta,mu
+; CHECK-NEXT:    vfsqrt.v v8, v8
+; CHECK-NEXT:    ret
+  %r = call <vscale x 8 x double> @llvm.sqrt.nxv8f64(<vscale x 8 x double> %v)
+  ret <vscale x 8 x double> %r
+}