test/CodeGen/AMDGPU/fsqrt.f64.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
; RUN: llc -global-isel=0 -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s

; RUN: llc -global-isel=1 -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
; RUN: llc -global-isel=1 -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s

define double @v_sqrt_f64(double %x) {
; GCN-LABEL: v_sqrt_f64:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_sqrt_f64_e32 v[0:1], v[0:1]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %result = call double @llvm.sqrt.f64(double %x)
  ret double %result
}

define double @v_sqrt_f64_fneg(double %x) {
; GCN-LABEL: v_sqrt_f64_fneg:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_sqrt_f64_e64 v[0:1], -v[0:1]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %x.neg = fneg double %x
  %result = call double @llvm.sqrt.f64(double %x.neg)
  ret double %result
}

define double @v_sqrt_f64_fabs(double %x) {
; GCN-LABEL: v_sqrt_f64_fabs:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_sqrt_f64_e64 v[0:1], |v[0:1]|
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %x.fabs = call double @llvm.fabs.f64(double %x)
  %result = call double @llvm.sqrt.f64(double %x.fabs)
  ret double %result
}

define double @v_sqrt_f64_fneg_fabs(double %x) {
; GCN-LABEL: v_sqrt_f64_fneg_fabs:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_sqrt_f64_e64 v[0:1], -|v[0:1]|
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %x.fabs = call double @llvm.fabs.f64(double %x)
  %x.fabs.neg = fneg double %x.fabs
  %result = call double @llvm.sqrt.f64(double %x.fabs.neg)
  ret double %result
}

define double @v_sqrt_f64_ninf(double %x) {
; GCN-LABEL: v_sqrt_f64_ninf:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_sqrt_f64_e32 v[0:1], v[0:1]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %result = call ninf double @llvm.sqrt.f64(double %x)
  ret double %result
}

define double @v_sqrt_f64_no_infs_attribute(double %x) "no-infs-fp-math"="true" {
; GCN-LABEL: v_sqrt_f64_no_infs_attribute:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_sqrt_f64_e32 v[0:1], v[0:1]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %result = call ninf double @llvm.sqrt.f64(double %x)
  ret double %result
}

define double @v_sqrt_f64_nnan(double %x) {
; GCN-LABEL: v_sqrt_f64_nnan:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_sqrt_f64_e32 v[0:1], v[0:1]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %result = call nnan double @llvm.sqrt.f64(double %x)
  ret double %result
}

define amdgpu_ps <2 x i32> @s_sqrt_f64(double inreg %x) {
; GCN-LABEL: s_sqrt_f64:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_sqrt_f64_e32 v[0:1], s[0:1]
; GCN-NEXT:    v_readfirstlane_b32 s0, v0
; GCN-NEXT:    v_readfirstlane_b32 s1, v1
; GCN-NEXT:    ; return to shader part epilog
  %result = call double @llvm.sqrt.f64(double %x)
  %cast = bitcast double %result to <2 x i32>
  %cast.0 = extractelement <2 x i32> %cast, i32 0
  %cast.1 = extractelement <2 x i32> %cast, i32 1
  %lane.0 = call i32 @llvm.amdgcn.readfirstlane(i32 %cast.0)
  %lane.1 = call i32 @llvm.amdgcn.readfirstlane(i32 %cast.1)
  %insert.0 = insertelement <2 x i32> poison, i32 %lane.0, i32 0
  %insert.1 = insertelement <2 x i32> %insert.0, i32 %lane.1, i32 1
  ret <2 x i32> %insert.1
}

define amdgpu_ps <2 x i32> @s_sqrt_f64_ninf(double inreg %x) {
; GCN-LABEL: s_sqrt_f64_ninf:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_sqrt_f64_e32 v[0:1], s[0:1]
; GCN-NEXT:    v_readfirstlane_b32 s0, v0
; GCN-NEXT:    v_readfirstlane_b32 s1, v1
; GCN-NEXT:    ; return to shader part epilog
  %result = call ninf double @llvm.sqrt.f64(double %x)
  %cast = bitcast double %result to <2 x i32>
  %cast.0 = extractelement <2 x i32> %cast, i32 0
  %cast.1 = extractelement <2 x i32> %cast, i32 1
  %lane.0 = call i32 @llvm.amdgcn.readfirstlane(i32 %cast.0)
  %lane.1 = call i32 @llvm.amdgcn.readfirstlane(i32 %cast.1)
  %insert.0 = insertelement <2 x i32> poison, i32 %lane.0, i32 0
  %insert.1 = insertelement <2 x i32> %insert.0, i32 %lane.1, i32 1
  ret <2 x i32> %insert.1
}

declare double @llvm.fabs.f64(double) #0
declare double @llvm.sqrt.f64(double) #0
declare i32 @llvm.amdgcn.readfirstlane(i32) #1

attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
attributes #1 = { convergent nounwind willreturn memory(none) }