[llvm] 11e0c5b - [AArch64] add another test for reciprocal sqrt; NFC

[Sanjay Patel via llvm-commits]

Author: Sanjay Patel
Date: 2020-08-31T09:35:59-04:00
New Revision: 11e0c5b648460256176f2348d724338218542dce

URL: https://github.com/llvm/llvm-project/commit/11e0c5b648460256176f2348d724338218542dce
DIFF: https://github.com/llvm/llvm-project/commit/11e0c5b648460256176f2348d724338218542dce.diff

LOG: [AArch64] add another test for reciprocal sqrt; NFC

Added: 
    

Modified: 
    llvm/test/CodeGen/AArch64/sqrt-fastmath.ll

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/AArch64/sqrt-fastmath.ll b/llvm/test/CodeGen/AArch64/sqrt-fastmath.ll
index ccc100e2b86f..85886938a186 100644
--- a/llvm/test/CodeGen/AArch64/sqrt-fastmath.ll
+++ b/llvm/test/CodeGen/AArch64/sqrt-fastmath.ll
@@ -567,6 +567,57 @@ define double @sqrt_simplify_before_recip_3_uses(double %x, double* %p1, double*
   ret double %sqrt_fast
 }
 
+define double @sqrt_simplify_before_recip_3_uses_order(double %x, double* %p1, double* %p2) nounwind {
+; FAULT-LABEL: sqrt_simplify_before_recip_3_uses_order:
+; FAULT:       // %bb.0:
+; FAULT-NEXT:    mov x8, #4631107791820423168
+; FAULT-NEXT:    fmov d3, x8
+; FAULT-NEXT:    mov x8, #140737488355328
+; FAULT-NEXT:    fsqrt d1, d0
+; FAULT-NEXT:    fmov d2, #1.00000000
+; FAULT-NEXT:    movk x8, #16453, lsl #48
+; FAULT-NEXT:    fdiv d1, d2, d1
+; FAULT-NEXT:    fmov d2, x8
+; FAULT-NEXT:    fmul d0, d0, d1
+; FAULT-NEXT:    fmul d3, d1, d3
+; FAULT-NEXT:    fmul d1, d1, d2
+; FAULT-NEXT:    str d3, [x0]
+; FAULT-NEXT:    str d1, [x1]
+; FAULT-NEXT:    ret
+;
+; CHECK-LABEL: sqrt_simplify_before_recip_3_uses_order:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    frsqrte d1, d0
+; CHECK-NEXT:    fmul d3, d1, d1
+; CHECK-NEXT:    frsqrts d3, d0, d3
+; CHECK-NEXT:    fmul d1, d1, d3
+; CHECK-NEXT:    fmul d3, d1, d1
+; CHECK-NEXT:    frsqrts d3, d0, d3
+; CHECK-NEXT:    mov x8, #4631107791820423168
+; CHECK-NEXT:    fmul d1, d1, d3
+; CHECK-NEXT:    fmov d2, x8
+; CHECK-NEXT:    mov x8, #140737488355328
+; CHECK-NEXT:    fmul d3, d1, d1
+; CHECK-NEXT:    movk x8, #16453, lsl #48
+; CHECK-NEXT:    frsqrts d3, d0, d3
+; CHECK-NEXT:    fmul d1, d1, d3
+; CHECK-NEXT:    fmov d3, x8
+; CHECK-NEXT:    fmul d0, d0, d1
+; CHECK-NEXT:    fmul d2, d1, d2
+; CHECK-NEXT:    fmul d1, d1, d3
+; CHECK-NEXT:    str d2, [x0]
+; CHECK-NEXT:    str d1, [x1]
+; CHECK-NEXT:    ret
+  %sqrt = tail call fast double @llvm.sqrt.f64(double %x)
+  %sqrt_fast = fdiv fast double %x, %sqrt
+  %r1 = fdiv fast double 42.0, %sqrt
+  %r2 = fdiv fast double 43.0, %sqrt
+  store double %r1, double* %p1, align 8
+  store double %r2, double* %p2, align 8
+  ret double %sqrt_fast
+}
+
+
 define double @sqrt_simplify_before_recip_4_uses(double %x, double* %p1, double* %p2, double* %p3) nounwind {
 ; FAULT-LABEL: sqrt_simplify_before_recip_4_uses:
 ; FAULT:       // %bb.0: