[AArch64] Don't map llvm sqrt intrinsics to veclib functions

[huntergr]

Summary

Since AArch64 has sqrt instructions, we want to use those instead of calls to vector math routines for llvm sqrt intrinsics (since those don't imply some of the constraints that libm calls might have) so we just remove the mappings.

Code originally written by @mgabka

[danielkiss]

LGTM

[paulwalker-arm]

I suspect we'll want to replace replace-intrinsics-with-veclib-sqrt.ll with more exhaustive testing for all the intrinsics but that's a job for another day. I've a couple of observations but otherwise the patch looks good.

llvm/test/CodeGen/AArch64/replace-intrinsics-with-veclib-sqrt.ll

2

Given the simplicity of the tests I doubt you need these extra arguments.

5

Is this required?

54–56

These three are not used within the test and can be removed?

llvm/include/llvm/Analysis/VecFuncs.def

 
 TLI_DEFINE_VECFUNC( "sin", "_ZGVnN2v_sin", FIXED(2))
 TLI_DEFINE_VECFUNC( "llvm.sin.f64", "_ZGVnN2v_sin", FIXED(2))
 
 TLI_DEFINE_VECFUNC( "sinh", "_ZGVnN2v_sinh", FIXED(2))
 TLI_DEFINE_VECFUNC( "llvm.sinh.f64", "_ZGVnN2v_sinh", FIXED(2))
 
 TLI_DEFINE_VECFUNC( "sqrt", "_ZGVnN2v_sqrt", FIXED(2))
-TLI_DEFINE_VECFUNC( "llvm.sqrt.f64", "_ZGVnN2v_sqrt", FIXED(2))
 
 TLI_DEFINE_VECFUNC( "tan", "_ZGVnN2v_tan", FIXED(2))
 TLI_DEFINE_VECFUNC( "llvm.tan.f64", "_ZGVnN2v_tan", FIXED(2))
 
 TLI_DEFINE_VECFUNC( "tanh", "_ZGVnN2v_tanh", FIXED(2))
 TLI_DEFINE_VECFUNC( "llvm.tanh.f64", "_ZGVnN2v_tanh", FIXED(2))
 
 TLI_DEFINE_VECFUNC( "tgamma", "_ZGVnN2v_tgamma", FIXED(2))

 
 TLI_DEFINE_VECFUNC( "sinf", "_ZGVnN4v_sinf", FIXED(4))
 TLI_DEFINE_VECFUNC( "llvm.sin.f32", "_ZGVnN4v_sinf", FIXED(4))
 
 TLI_DEFINE_VECFUNC( "sinhf", "_ZGVnN4v_sinhf", FIXED(4))
 TLI_DEFINE_VECFUNC( "llvm.sinh.f32", "_ZGVnN4v_sinhf", FIXED(4))
 
 TLI_DEFINE_VECFUNC( "sqrtf", "_ZGVnN4v_sqrtf", FIXED(4))
-TLI_DEFINE_VECFUNC( "llvm.sqrt.f32", "_ZGVnN4v_sqrtf", FIXED(4))
 
 TLI_DEFINE_VECFUNC( "tanf", "_ZGVnN4v_tanf", FIXED(4))
 TLI_DEFINE_VECFUNC( "llvm.tan.f32", "_ZGVnN4v_tanf", FIXED(4))
 
 TLI_DEFINE_VECFUNC( "tanhf", "_ZGVnN4v_tanhf", FIXED(4))
 TLI_DEFINE_VECFUNC( "llvm.tanh.f32", "_ZGVnN4v_tanhf", FIXED(4))
 
 TLI_DEFINE_VECFUNC( "tgammaf", "_ZGVnN4v_tgammaf", FIXED(4))

llvm/test/CodeGen/AArch64/replace-intrinsics-with-veclib-sqrt.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -vector-library=sleefgnuabi -replace-with-veclib -S < %s | FileCheck %s
+
+target triple = "aarch64-unknown-linux-gnu"
+
+define <2 x double> @llvm_sqrt_f64(<2 x double> %in) {
+; CHECK-LABEL: define {{[^@]+}}@llvm_sqrt_f64
+; CHECK-SAME: (<2 x double> [[IN:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[IN]])
+; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+;
+  %1 = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> %in)
+  ret <2 x double> %1
+}
+
+define <4 x float> @llvm_sqrt_f32(<4 x float> %in) {
+; CHECK-LABEL: define {{[^@]+}}@llvm_sqrt_f32
+; CHECK-SAME: (<4 x float> [[IN:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = call fast <4 x float> @llvm.sqrt.v4f32(<4 x float> [[IN]])
+; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+;
+  %1 = call fast <4 x float> @llvm.sqrt.v4f32(<4 x float> %in)
+  ret <4 x float> %1
+}
+
+define <vscale x 2 x double> @llvm_sqrt_vscale_f64(<vscale x 2 x double> %in) {
+; CHECK-LABEL: define {{[^@]+}}@llvm_sqrt_vscale_f64
+; CHECK-SAME: (<vscale x 2 x double> [[IN:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = call fast <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double> [[IN]])
+; CHECK-NEXT:    ret <vscale x 2 x double> [[TMP1]]
+;
+  %1 = call fast <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double> %in)
+  ret <vscale x 2 x double> %1
+}
+
+define <vscale x 4 x float> @llvm_sqrt_vscale_f32(<vscale x 4 x float> %in) {
+; CHECK-LABEL: define {{[^@]+}}@llvm_sqrt_vscale_f32
+; CHECK-SAME: (<vscale x 4 x float> [[IN:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = call fast <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float> [[IN]])
+; CHECK-NEXT:    ret <vscale x 4 x float> [[TMP1]]
+;
+  %1 = call fast <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float> %in)
+  ret <vscale x 4 x float> %1
+}
+
+
+declare <2 x double> @llvm.sqrt.v2f64(<2 x double>)
+declare <4 x float> @llvm.sqrt.v4f32(<4 x float>)
+declare <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double>)
+declare <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float>)

llvm/test/Transforms/LoopVectorize/AArch64/sleef-calls-aarch64.ll

define void @sqrt_f32(float* nocapture %varray) {
   %iv.next = add nuw nsw i64 %iv, 1
   %exitcond = icmp eq i64 %iv.next, 1000
   br i1 %exitcond, label %for.end, label %for.body
 
   for.end:
   ret void
 }
 
+  
+define void @llvm_sqrt_f64(double* nocapture %varray) {
+  ; CHECK-LABEL: @llvm_sqrt_f64(
+  ; CHECK:    [[TMP5:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[TMP4:%.*]])
+  ; CHECK:    ret void
+  ;
+  entry:
+  br label %for.body
+
+  for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call fast double @llvm.sqrt.f64(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %iv
+  store double %call, double* %arrayidx, align 8
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+  for.end:
+  ret void
+}
+
+define void @llvm_sqrt_f32(float* nocapture %varray) {
+  ; CHECK-LABEL: @llvm_sqrt_f32(
+  ; CHECK:    [[TMP5:%.*]] = call fast <4 x float> @llvm.sqrt.v4f32(<4 x float> [[TMP4:%.*]])
+  ; CHECK:    ret void
+  ;
+  entry:
+  br label %for.body
+
+  for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call fast float @llvm.sqrt.f32(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %iv
+  store float %call, float* %arrayidx, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+  for.end:
+  ret void
+}
+
 declare double @tan(double) #0
 declare float @tanf(float) #0
 declare double @llvm.tan.f64(double) #0
 declare float @llvm.tan.f32(float) #0
 
 define void @tan_f64(double* nocapture %varray) {
   ; CHECK-LABEL: @tan_f64(
   ; CHECK:    [[TMP5:%.*]] = call <2 x double> @_ZGVnN2v_tan(<2 x double> [[TMP4:%.*]])