llvm/test/CodeGen/Hexagon/bfloat.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=hexagon -hexagon-bit=false < %s | FileCheck %s

define bfloat @load_scalar_bf(ptr %addr) {
; CHECK-LABEL: load_scalar_bf:
; CHECK:         .cfi_startproc
; CHECK-NEXT:  // %bb.0: // %entry
; CHECK-NEXT:    {
; CHECK-NEXT:     r0 = memuh(r0+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r0 = aslh(r0)
; CHECK-NEXT:     jumpr r31
; CHECK-NEXT:    }

entry:
  %0 = load bfloat, ptr %addr, align 2
  ret bfloat %0
}

define void @store_scalar_bf(bfloat %v, ptr %addr) {
; CHECK-LABEL: store_scalar_bf:
; CHECK:         .cfi_startproc
; CHECK-NEXT:  // %bb.0: // %entry
; CHECK-NEXT:    {
; CHECK-NEXT:     r0 = lsr(r0,#16)
; CHECK-NEXT:     jumpr r31
; CHECK-NEXT:     memh(r1+#0) = r0.new
; CHECK-NEXT:    }

entry:
  store bfloat %v, ptr %addr, align 2
  ret void
}

define bfloat @sum(bfloat %a, bfloat %b) #0 {
; CHECK-LABEL: sum:
; CHECK:         .cfi_startproc
; CHECK-NEXT:  // %bb.0: // %entry
; CHECK-NEXT:    {
; CHECK-NEXT:     r1 = and(r1,##-65536)
; CHECK-NEXT:     r2 = ##32768
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r3 = ##131071
; CHECK-NEXT:     r4 = ##65535
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r0 = insert(r1,#16,#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r0 = sfadd(r0,r1)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r1 = and(r0,r2)
; CHECK-NEXT:     r3 = and(r0,r3)
; CHECK-NEXT:     r5 = asrh(r0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r1 = add(r0,r1)
; CHECK-NEXT:     p0 = cmp.eq(r3,r2)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r1 = asrh(r1)
; CHECK-NEXT:     if (p0) r2 = and(r5,r4)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     p0 = sfclass(r0,#16)
; CHECK-NEXT:     if (!p0) r2 = and(r1,r4)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     if (p0) r2 = ##32767
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r0 = aslh(r2)
; CHECK-NEXT:     jumpr r31
; CHECK-NEXT:    }

entry:
  %add = fadd bfloat %a, %b
  ret bfloat %add
}

; Function Attrs: noinline nounwind optnone
define dso_local i32 @double_bf16(bfloat %a, bfloat %b) #0 {
; CHECK-LABEL: double_bf16:
; CHECK:         .cfi_startproc
; CHECK-NEXT:  // %bb.0: // %entry
; CHECK-NEXT:    {
; CHECK-NEXT:     r3:2 = convert_sf2df(r0)
; CHECK-NEXT:     r5:4 = convert_sf2df(r1)
; CHECK-NEXT:     r29 = add(r29,#-16)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r1:0 = dfadd(r3:2,r5:4)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     memd(r29+#8) = r1:0
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r2 = convert_df2sf(r1:0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r1 = lsr(r2,#16)
; CHECK-NEXT:     r0 = and(r2,##-65536)
; CHECK-NEXT:     memh(r29+#6) = r1.new
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r0 = convert_sf2w(r0):chop
; CHECK-NEXT:     jumpr r31
; CHECK-NEXT:     r29 = add(r29,#16)
; CHECK-NEXT:    }

entry:
  %c = alloca double, align 8
  %d = alloca bfloat, align 2
  %conv = fpext bfloat %a to double
  %conv1 = fpext bfloat %b to double
  %add = fadd double %conv, %conv1
  store double %add, ptr %c, align 8
  %2 = load double, ptr %c, align 8
  %conv2 = fptrunc double %2 to bfloat
  store bfloat %conv2, ptr %d, align 2
  %3 = load bfloat, ptr %d, align 2
  %conv3 = fptosi bfloat %3 to i32
  ret i32 %conv3
}