5060f08c3a
Currently bf16 fpextend is lowered to a vector shift. Instead leave it as fpextend and have an instruction selection pattern which selects to a shift later. Doing this means that DAGCombiner patterns for fpextend will be applied, leading to better codegen. It also means that in some situations we use a mov instruction where we previously have a dup instruction, but I don't think this makes any difference.
2270 lines
83 KiB
LLVM
2270 lines
83 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc < %s -mtriple aarch64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,CHECK-CVT
|
|
; RUN: llc < %s -mtriple aarch64-unknown-unknown -mattr=+bf16 | FileCheck %s --check-prefixes=CHECK,CHECK-BF16,CHECK-SD
|
|
; RUN: llc < %s -mtriple aarch64-unknown-unknown -mattr=+bf16,+fullfp16 -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-BF16,CHECK-GI
|
|
|
|
; CHECK-GI: warning: Instruction selection used fallback path for test_fadd
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fsub
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fmul
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fmadd
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fdiv
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_frem
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_call
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_call_flipped
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_tailcall_flipped
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_select_cc
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_select_cc_f32_f16
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_une
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_ueq
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_ugt
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_uge
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_ult
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_ule
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_uno
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_one
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_oeq
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_ogt
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_oge
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_olt
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_ole
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fcmp_ord
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fccmp
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_br_cc
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fptosi_i32
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fptosi_i64
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fptoui_i32
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fptoui_i64
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_uitofp_i32
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_uitofp_i64
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_sitofp_i32
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_sitofp_i64
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_uitofp_i32_fadd
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_sitofp_i32_fadd
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fptrunc_float
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fptrunc_double
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fpext_float
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fpext_double
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_sqrt
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_powi
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_sin
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_cos
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_tan
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_acos
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_asin
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_atan
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_atan2
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_cosh
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_sinh
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_tanh
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_pow
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_exp
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_exp2
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_log
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_log10
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_log2
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fma
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fabs
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_minnum
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_maxnum
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_copysign
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_copysign_f32
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_copysign_f64
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_copysign_extended
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_floor
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_ceil
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_trunc
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_rint
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_nearbyint
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_round
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_roundeven
|
|
; CHECK-GI-NEXT: warning: Instruction selection used fallback path for test_fmuladd
|
|
|
|
define bfloat @test_fadd(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_fadd:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fadd s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fadd:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fadd s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = fadd bfloat %a, %b
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fsub(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_fsub:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fsub s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fsub:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fsub s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = fsub bfloat %a, %b
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fmul(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_fmul:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fmul s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fmul:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fmul s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = fmul bfloat %a, %b
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fmadd(bfloat %a, bfloat %b, bfloat %c) #0 {
|
|
; CHECK-CVT-LABEL: test_fmadd:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w10, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fmul s0, s0, s1
|
|
; CHECK-CVT-NEXT: shll v1.4s, v2.4h, #16
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: ubfx w9, w8, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w8, w10
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fadd s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: ubfx w9, w8, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w8, w10
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fmadd:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: shll v2.4s, v2.4h, #16
|
|
; CHECK-BF16-NEXT: fmadd s0, s0, s1, s2
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%mul = fmul fast bfloat %a, %b
|
|
%r = fadd fast bfloat %mul, %c
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fdiv(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_fdiv:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fdiv s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fdiv:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fdiv s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = fdiv bfloat %a, %b
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_frem(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_frem:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: // kill: def $s1 killed $s1 killed $q1
|
|
; CHECK-CVT-NEXT: bl fmodf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_frem:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: // kill: def $s1 killed $s1 killed $q1
|
|
; CHECK-BF16-NEXT: bl fmodf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = frem bfloat %a, %b
|
|
ret bfloat %r
|
|
}
|
|
|
|
define void @test_store(bfloat %a, ptr %b) #0 {
|
|
; CHECK-LABEL: test_store:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: str h0, [x0]
|
|
; CHECK-NEXT: ret
|
|
store bfloat %a, ptr %b
|
|
ret void
|
|
}
|
|
|
|
define bfloat @test_load(ptr %a) #0 {
|
|
; CHECK-LABEL: test_load:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldr h0, [x0]
|
|
; CHECK-NEXT: ret
|
|
%r = load bfloat, ptr %a
|
|
ret bfloat %r
|
|
}
|
|
|
|
declare bfloat @test_callee(bfloat %a, bfloat %b) #0
|
|
|
|
define bfloat @test_call(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_call:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-NEXT: bl test_callee
|
|
; CHECK-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-NEXT: ret
|
|
%r = call bfloat @test_callee(bfloat %a, bfloat %b)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_call_flipped(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_call_flipped:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-NEXT: fmov s2, s0
|
|
; CHECK-NEXT: fmov s0, s1
|
|
; CHECK-NEXT: fmov s1, s2
|
|
; CHECK-NEXT: bl test_callee
|
|
; CHECK-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-NEXT: ret
|
|
%r = call bfloat @test_callee(bfloat %b, bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_tailcall_flipped(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_tailcall_flipped:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: fmov s2, s0
|
|
; CHECK-NEXT: fmov s0, s1
|
|
; CHECK-NEXT: fmov s1, s2
|
|
; CHECK-NEXT: b test_callee
|
|
%r = tail call bfloat @test_callee(bfloat %b, bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_select(bfloat %a, bfloat %b, i1 zeroext %c) #0 {
|
|
; CHECK-CVT-LABEL: test_select:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-CVT-NEXT: cmp w0, #0
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $s1
|
|
; CHECK-CVT-NEXT: fcsel s0, s0, s1, ne
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-SD-LABEL: test_select:
|
|
; CHECK-SD: // %bb.0:
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-SD-NEXT: cmp w0, #0
|
|
; CHECK-SD-NEXT: // kill: def $h1 killed $h1 def $s1
|
|
; CHECK-SD-NEXT: fcsel s0, s0, s1, ne
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-SD-NEXT: ret
|
|
;
|
|
; CHECK-GI-LABEL: test_select:
|
|
; CHECK-GI: // %bb.0:
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-GI-NEXT: // kill: def $h1 killed $h1 def $s1
|
|
; CHECK-GI-NEXT: fmov w8, s0
|
|
; CHECK-GI-NEXT: fmov w9, s1
|
|
; CHECK-GI-NEXT: tst w0, #0x1
|
|
; CHECK-GI-NEXT: csel w8, w8, w9, ne
|
|
; CHECK-GI-NEXT: fmov s0, w8
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-GI-NEXT: ret
|
|
%r = select i1 %c, bfloat %a, bfloat %b
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_select_cc(bfloat %a, bfloat %b, bfloat %c, bfloat %d) #0 {
|
|
; CHECK-CVT-LABEL: test_select_cc:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h3 killed $h3 def $d3
|
|
; CHECK-CVT-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $s1
|
|
; CHECK-CVT-NEXT: shll v3.4s, v3.4h, #16
|
|
; CHECK-CVT-NEXT: shll v2.4s, v2.4h, #16
|
|
; CHECK-CVT-NEXT: fcmp s2, s3
|
|
; CHECK-CVT-NEXT: fcsel s0, s0, s1, ne
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-SD-LABEL: test_select_cc:
|
|
; CHECK-SD: // %bb.0:
|
|
; CHECK-SD-NEXT: // kill: def $h3 killed $h3 def $d3
|
|
; CHECK-SD-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-SD-NEXT: // kill: def $h1 killed $h1 def $s1
|
|
; CHECK-SD-NEXT: shll v3.4s, v3.4h, #16
|
|
; CHECK-SD-NEXT: shll v2.4s, v2.4h, #16
|
|
; CHECK-SD-NEXT: fcmp s2, s3
|
|
; CHECK-SD-NEXT: fcsel s0, s0, s1, ne
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-SD-NEXT: ret
|
|
;
|
|
; CHECK-GI-LABEL: test_select_cc:
|
|
; CHECK-GI: // %bb.0:
|
|
; CHECK-GI-NEXT: // kill: def $h3 killed $h3 def $d3
|
|
; CHECK-GI-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-GI-NEXT: shll v3.4s, v3.4h, #16
|
|
; CHECK-GI-NEXT: shll v2.4s, v2.4h, #16
|
|
; CHECK-GI-NEXT: fcmp s2, s3
|
|
; CHECK-GI-NEXT: fcsel h0, h0, h1, ne
|
|
; CHECK-GI-NEXT: ret
|
|
%cc = fcmp une bfloat %c, %d
|
|
%r = select i1 %cc, bfloat %a, bfloat %b
|
|
ret bfloat %r
|
|
}
|
|
|
|
define float @test_select_cc_f32_f16(float %a, float %b, bfloat %c, bfloat %d) #0 {
|
|
; CHECK-LABEL: test_select_cc_f32_f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h3 killed $h3 def $d3
|
|
; CHECK-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-NEXT: shll v3.4s, v3.4h, #16
|
|
; CHECK-NEXT: shll v2.4s, v2.4h, #16
|
|
; CHECK-NEXT: fcmp s2, s3
|
|
; CHECK-NEXT: fcsel s0, s0, s1, ne
|
|
; CHECK-NEXT: ret
|
|
%cc = fcmp une bfloat %c, %d
|
|
%r = select i1 %cc, float %a, float %b
|
|
ret float %r
|
|
}
|
|
|
|
define bfloat @test_select_cc_f16_f32(bfloat %a, bfloat %b, float %c, float %d) #0 {
|
|
; CHECK-CVT-LABEL: test_select_cc_f16_f32:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: fcmp s2, s3
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $s1
|
|
; CHECK-CVT-NEXT: fcsel s0, s0, s1, ne
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-SD-LABEL: test_select_cc_f16_f32:
|
|
; CHECK-SD: // %bb.0:
|
|
; CHECK-SD-NEXT: fcmp s2, s3
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-SD-NEXT: // kill: def $h1 killed $h1 def $s1
|
|
; CHECK-SD-NEXT: fcsel s0, s0, s1, ne
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-SD-NEXT: ret
|
|
;
|
|
; CHECK-GI-LABEL: test_select_cc_f16_f32:
|
|
; CHECK-GI: // %bb.0:
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-GI-NEXT: // kill: def $h1 killed $h1 def $s1
|
|
; CHECK-GI-NEXT: fcmp s2, s3
|
|
; CHECK-GI-NEXT: fmov w8, s0
|
|
; CHECK-GI-NEXT: fmov w9, s1
|
|
; CHECK-GI-NEXT: csel w8, w8, w9, ne
|
|
; CHECK-GI-NEXT: fmov s0, w8
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-GI-NEXT: ret
|
|
%cc = fcmp une float %c, %d
|
|
%r = select i1 %cc, bfloat %a, bfloat %b
|
|
ret bfloat %r
|
|
}
|
|
|
|
define i1 @test_fcmp_une(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_une:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, ne
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp une bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_ueq(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_ueq:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w8, eq
|
|
; CHECK-NEXT: csinc w0, w8, wzr, vc
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp ueq bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_ugt(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_ugt:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, hi
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp ugt bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_uge(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_uge:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, pl
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp uge bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_ult(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_ult:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, lt
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp ult bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_ule(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_ule:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, le
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp ule bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_uno(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_uno:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, vs
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp uno bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_one(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_one:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w8, mi
|
|
; CHECK-NEXT: csinc w0, w8, wzr, le
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp one bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_oeq(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_oeq:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, eq
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp oeq bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_ogt(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_ogt:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, gt
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp ogt bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_oge(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_oge:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, ge
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp oge bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_olt(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_olt:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, mi
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp olt bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_ole(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_ole:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, ls
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp ole bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define i1 @test_fcmp_ord(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-LABEL: test_fcmp_ord:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: cset w0, vc
|
|
; CHECK-NEXT: ret
|
|
%r = fcmp ord bfloat %a, %b
|
|
ret i1 %r
|
|
}
|
|
|
|
define void @test_fccmp(bfloat %in, ptr %out) {
|
|
; CHECK-CVT-LABEL: test_fccmp:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: movi v1.2s, #69, lsl #24
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v2.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: adrp x8, .LCPI29_0
|
|
; CHECK-CVT-NEXT: movi v3.2s, #72, lsl #24
|
|
; CHECK-CVT-NEXT: fcmp s2, s1
|
|
; CHECK-CVT-NEXT: ldr h1, [x8, :lo12:.LCPI29_0]
|
|
; CHECK-CVT-NEXT: fccmp s2, s3, #4, mi
|
|
; CHECK-CVT-NEXT: fcsel s0, s0, s1, gt
|
|
; CHECK-CVT-NEXT: str h0, [x0]
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-SD-LABEL: test_fccmp:
|
|
; CHECK-SD: // %bb.0:
|
|
; CHECK-SD-NEXT: movi v1.2s, #69, lsl #24
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-SD-NEXT: shll v2.4s, v0.4h, #16
|
|
; CHECK-SD-NEXT: adrp x8, .LCPI29_0
|
|
; CHECK-SD-NEXT: movi v3.2s, #72, lsl #24
|
|
; CHECK-SD-NEXT: fcmp s2, s1
|
|
; CHECK-SD-NEXT: ldr h1, [x8, :lo12:.LCPI29_0]
|
|
; CHECK-SD-NEXT: fccmp s2, s3, #4, mi
|
|
; CHECK-SD-NEXT: fcsel s0, s0, s1, gt
|
|
; CHECK-SD-NEXT: str h0, [x0]
|
|
; CHECK-SD-NEXT: ret
|
|
;
|
|
; CHECK-GI-LABEL: test_fccmp:
|
|
; CHECK-GI: // %bb.0:
|
|
; CHECK-GI-NEXT: movi v1.2s, #69, lsl #24
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-GI-NEXT: shll v2.4s, v0.4h, #16
|
|
; CHECK-GI-NEXT: movi v3.2s, #72, lsl #24
|
|
; CHECK-GI-NEXT: fcmp s2, s1
|
|
; CHECK-GI-NEXT: fmov h1, #5.00000000
|
|
; CHECK-GI-NEXT: fccmp s2, s3, #4, mi
|
|
; CHECK-GI-NEXT: fcsel h0, h0, h1, gt
|
|
; CHECK-GI-NEXT: str h0, [x0]
|
|
; CHECK-GI-NEXT: ret
|
|
%cmp1 = fcmp ogt bfloat %in, 0xR4800
|
|
%cmp2 = fcmp olt bfloat %in, 0xR4500
|
|
%cond = and i1 %cmp1, %cmp2
|
|
%result = select i1 %cond, bfloat %in, bfloat 0xR4500
|
|
store bfloat %result, ptr %out
|
|
ret void
|
|
}
|
|
|
|
define void @test_br_cc(bfloat %a, bfloat %b, ptr %p1, ptr %p2) #0 {
|
|
; CHECK-LABEL: test_br_cc:
|
|
; CHECK: // %bb.0: // %common.ret
|
|
; CHECK-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcmp s0, s1
|
|
; CHECK-NEXT: csel x8, x0, x1, pl
|
|
; CHECK-NEXT: str wzr, [x8]
|
|
; CHECK-NEXT: ret
|
|
%c = fcmp uge bfloat %a, %b
|
|
br i1 %c, label %then, label %else
|
|
then:
|
|
store i32 0, ptr %p1
|
|
ret void
|
|
else:
|
|
store i32 0, ptr %p2
|
|
ret void
|
|
}
|
|
|
|
define bfloat @test_phi(ptr %p1) #0 {
|
|
; CHECK-LABEL: test_phi:
|
|
; CHECK: // %bb.0: // %entry
|
|
; CHECK-NEXT: stp d9, d8, [sp, #-32]! // 16-byte Folded Spill
|
|
; CHECK-NEXT: ldr h9, [x0]
|
|
; CHECK-NEXT: stp x30, x19, [sp, #16] // 16-byte Folded Spill
|
|
; CHECK-NEXT: mov x19, x0
|
|
; CHECK-NEXT: .LBB31_1: // %loop
|
|
; CHECK-NEXT: // =>This Inner Loop Header: Depth=1
|
|
; CHECK-NEXT: fmov s8, s9
|
|
; CHECK-NEXT: ldr h9, [x19]
|
|
; CHECK-NEXT: mov x0, x19
|
|
; CHECK-NEXT: bl test_dummy
|
|
; CHECK-NEXT: tbnz w0, #0, .LBB31_1
|
|
; CHECK-NEXT: // %bb.2: // %return
|
|
; CHECK-NEXT: ldp x30, x19, [sp, #16] // 16-byte Folded Reload
|
|
; CHECK-NEXT: fmov s0, s8
|
|
; CHECK-NEXT: ldp d9, d8, [sp], #32 // 16-byte Folded Reload
|
|
; CHECK-NEXT: ret
|
|
entry:
|
|
%a = load bfloat, ptr %p1
|
|
br label %loop
|
|
loop:
|
|
%r = phi bfloat [%a, %entry], [%b, %loop]
|
|
%b = load bfloat, ptr %p1
|
|
%c = call i1 @test_dummy(ptr %p1)
|
|
br i1 %c, label %loop, label %return
|
|
return:
|
|
ret bfloat %r
|
|
}
|
|
|
|
declare i1 @test_dummy(ptr %p1) #0
|
|
|
|
define i32 @test_fptosi_i32(bfloat %a) #0 {
|
|
; CHECK-LABEL: test_fptosi_i32:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcvtzs w0, s0
|
|
; CHECK-NEXT: ret
|
|
%r = fptosi bfloat %a to i32
|
|
ret i32 %r
|
|
}
|
|
|
|
define i64 @test_fptosi_i64(bfloat %a) #0 {
|
|
; CHECK-LABEL: test_fptosi_i64:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcvtzs x0, s0
|
|
; CHECK-NEXT: ret
|
|
%r = fptosi bfloat %a to i64
|
|
ret i64 %r
|
|
}
|
|
|
|
define i32 @test_fptoui_i32(bfloat %a) #0 {
|
|
; CHECK-LABEL: test_fptoui_i32:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcvtzu w0, s0
|
|
; CHECK-NEXT: ret
|
|
%r = fptoui bfloat %a to i32
|
|
ret i32 %r
|
|
}
|
|
|
|
define i64 @test_fptoui_i64(bfloat %a) #0 {
|
|
; CHECK-LABEL: test_fptoui_i64:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcvtzu x0, s0
|
|
; CHECK-NEXT: ret
|
|
%r = fptoui bfloat %a to i64
|
|
ret i64 %r
|
|
}
|
|
|
|
define bfloat @test_uitofp_i32(i32 %a) #0 {
|
|
; CHECK-CVT-LABEL: test_uitofp_i32:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: ucvtf d0, w0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: fcvtxn s0, d0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_uitofp_i32:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: ucvtf d0, w0
|
|
; CHECK-BF16-NEXT: fcvtxn s0, d0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = uitofp i32 %a to bfloat
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_uitofp_i64(i64 %a) #0 {
|
|
; CHECK-CVT-LABEL: test_uitofp_i64:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: lsr x9, x0, #53
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: cmp x9, #0
|
|
; CHECK-CVT-NEXT: and x9, x0, #0xfffffffffffff000
|
|
; CHECK-CVT-NEXT: csel x9, x9, x0, ne
|
|
; CHECK-CVT-NEXT: ucvtf d0, x9
|
|
; CHECK-CVT-NEXT: cset w9, ne
|
|
; CHECK-CVT-NEXT: tst x0, #0xfff
|
|
; CHECK-CVT-NEXT: csel w9, wzr, w9, eq
|
|
; CHECK-CVT-NEXT: fmov x10, d0
|
|
; CHECK-CVT-NEXT: orr x9, x10, x9
|
|
; CHECK-CVT-NEXT: fmov d0, x9
|
|
; CHECK-CVT-NEXT: fcvtxn s0, d0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_uitofp_i64:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: lsr x8, x0, #53
|
|
; CHECK-BF16-NEXT: and x9, x0, #0xfffffffffffff000
|
|
; CHECK-BF16-NEXT: cmp x8, #0
|
|
; CHECK-BF16-NEXT: csel x8, x9, x0, ne
|
|
; CHECK-BF16-NEXT: ucvtf d0, x8
|
|
; CHECK-BF16-NEXT: cset w8, ne
|
|
; CHECK-BF16-NEXT: tst x0, #0xfff
|
|
; CHECK-BF16-NEXT: csel w8, wzr, w8, eq
|
|
; CHECK-BF16-NEXT: fmov x9, d0
|
|
; CHECK-BF16-NEXT: orr x8, x9, x8
|
|
; CHECK-BF16-NEXT: fmov d0, x8
|
|
; CHECK-BF16-NEXT: fcvtxn s0, d0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = uitofp i64 %a to bfloat
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_sitofp_i32(i32 %a) #0 {
|
|
; CHECK-CVT-LABEL: test_sitofp_i32:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: scvtf d0, w0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: fcvtxn s0, d0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_sitofp_i32:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: scvtf d0, w0
|
|
; CHECK-BF16-NEXT: fcvtxn s0, d0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = sitofp i32 %a to bfloat
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_sitofp_i64(i64 %a) #0 {
|
|
; CHECK-CVT-LABEL: test_sitofp_i64:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: cmp x0, #0
|
|
; CHECK-CVT-NEXT: and x11, x0, #0x8000000000000000
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: cneg x9, x0, mi
|
|
; CHECK-CVT-NEXT: lsr x10, x9, #53
|
|
; CHECK-CVT-NEXT: cmp x10, #0
|
|
; CHECK-CVT-NEXT: and x10, x9, #0xfffffffffffff000
|
|
; CHECK-CVT-NEXT: csel x10, x10, x9, ne
|
|
; CHECK-CVT-NEXT: scvtf d0, x10
|
|
; CHECK-CVT-NEXT: cset w10, ne
|
|
; CHECK-CVT-NEXT: tst x9, #0xfff
|
|
; CHECK-CVT-NEXT: csel w10, wzr, w10, eq
|
|
; CHECK-CVT-NEXT: fmov x9, d0
|
|
; CHECK-CVT-NEXT: orr x9, x9, x11
|
|
; CHECK-CVT-NEXT: orr x9, x9, x10
|
|
; CHECK-CVT-NEXT: fmov d0, x9
|
|
; CHECK-CVT-NEXT: fcvtxn s0, d0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_sitofp_i64:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: cmp x0, #0
|
|
; CHECK-BF16-NEXT: cneg x8, x0, mi
|
|
; CHECK-BF16-NEXT: lsr x9, x8, #53
|
|
; CHECK-BF16-NEXT: and x10, x8, #0xfffffffffffff000
|
|
; CHECK-BF16-NEXT: cmp x9, #0
|
|
; CHECK-BF16-NEXT: csel x9, x10, x8, ne
|
|
; CHECK-BF16-NEXT: and x10, x0, #0x8000000000000000
|
|
; CHECK-BF16-NEXT: cset w11, ne
|
|
; CHECK-BF16-NEXT: scvtf d0, x9
|
|
; CHECK-BF16-NEXT: tst x8, #0xfff
|
|
; CHECK-BF16-NEXT: fmov x9, d0
|
|
; CHECK-BF16-NEXT: orr x8, x9, x10
|
|
; CHECK-BF16-NEXT: csel w9, wzr, w11, eq
|
|
; CHECK-BF16-NEXT: orr x8, x8, x9
|
|
; CHECK-BF16-NEXT: fmov d0, x8
|
|
; CHECK-BF16-NEXT: fcvtxn s0, d0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = sitofp i64 %a to bfloat
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_uitofp_i32_fadd(i32 %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_uitofp_i32_fadd:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: ucvtf d1, w0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fcvtxn s1, d1
|
|
; CHECK-CVT-NEXT: fmov w9, s1
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w9, w9, w8
|
|
; CHECK-CVT-NEXT: add w9, w10, w9
|
|
; CHECK-CVT-NEXT: lsr w9, w9, #16
|
|
; CHECK-CVT-NEXT: fmov s1, w9
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: fadd s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_uitofp_i32_fadd:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: ucvtf d1, w0
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fcvtxn s1, d1
|
|
; CHECK-BF16-NEXT: bfcvt h1, s1
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: fadd s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%c = uitofp i32 %a to bfloat
|
|
%r = fadd bfloat %b, %c
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_sitofp_i32_fadd(i32 %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_sitofp_i32_fadd:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: scvtf d1, w0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fcvtxn s1, d1
|
|
; CHECK-CVT-NEXT: fmov w9, s1
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w9, w9, w8
|
|
; CHECK-CVT-NEXT: add w9, w10, w9
|
|
; CHECK-CVT-NEXT: lsr w9, w9, #16
|
|
; CHECK-CVT-NEXT: fmov s1, w9
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: fadd s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_sitofp_i32_fadd:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: scvtf d1, w0
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fcvtxn s1, d1
|
|
; CHECK-BF16-NEXT: bfcvt h1, s1
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: fadd s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%c = sitofp i32 %a to bfloat
|
|
%r = fadd bfloat %b, %c
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fptrunc_float(float %a) #0 {
|
|
; CHECK-CVT-LABEL: test_fptrunc_float:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: fcmp s0, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: orr w9, w9, #0x400000
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: csel w8, w9, w8, vs
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fptrunc_float:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = fptrunc float %a to bfloat
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fptrunc_double(double %a) #0 {
|
|
; CHECK-CVT-LABEL: test_fptrunc_double:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: fcvtxn s0, d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fptrunc_double:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: fcvtxn s0, d0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = fptrunc double %a to bfloat
|
|
ret bfloat %r
|
|
}
|
|
|
|
define float @test_fpext_float(bfloat %a) #0 {
|
|
; CHECK-LABEL: test_fpext_float:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-NEXT: ret
|
|
%r = fpext bfloat %a to float
|
|
ret float %r
|
|
}
|
|
|
|
define double @test_fpext_double(bfloat %a) #0 {
|
|
; CHECK-LABEL: test_fpext_double:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-NEXT: fcvt d0, s0
|
|
; CHECK-NEXT: ret
|
|
%r = fpext bfloat %a to double
|
|
ret double %r
|
|
}
|
|
|
|
define i16 @test_bitcast_bfloattoi16(bfloat %a) #0 {
|
|
; CHECK-LABEL: test_bitcast_bfloattoi16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-NEXT: fmov w0, s0
|
|
; CHECK-NEXT: ret
|
|
%r = bitcast bfloat %a to i16
|
|
ret i16 %r
|
|
}
|
|
|
|
define bfloat @test_bitcast_i16tobfloat(i16 %a) #0 {
|
|
; CHECK-LABEL: test_bitcast_i16tobfloat:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: fmov s0, w0
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-NEXT: ret
|
|
%r = bitcast i16 %a to bfloat
|
|
ret bfloat %r
|
|
}
|
|
|
|
declare bfloat @llvm.sqrt.f16(bfloat %a) #0
|
|
declare bfloat @llvm.powi.f16.i32(bfloat %a, i32 %b) #0
|
|
declare bfloat @llvm.sin.f16(bfloat %a) #0
|
|
declare bfloat @llvm.cos.f16(bfloat %a) #0
|
|
declare bfloat @llvm.tan.f16(bfloat %a) #0
|
|
declare bfloat @llvm.asin.f16(bfloat %a) #0
|
|
declare bfloat @llvm.acos.f16(bfloat %a) #0
|
|
declare bfloat @llvm.atan.f16(bfloat %a) #0
|
|
declare bfloat @llvm.atan2.f16(bfloat %a, bfloat %b) #0
|
|
declare bfloat @llvm.sinh.f16(bfloat %a) #0
|
|
declare bfloat @llvm.cosh.f16(bfloat %a) #0
|
|
declare bfloat @llvm.tanh.f16(bfloat %a) #0
|
|
declare bfloat @llvm.pow.f16(bfloat %a, bfloat %b) #0
|
|
declare bfloat @llvm.exp.f16(bfloat %a) #0
|
|
declare bfloat @llvm.exp2.f16(bfloat %a) #0
|
|
declare bfloat @llvm.log.f16(bfloat %a) #0
|
|
declare bfloat @llvm.log10.f16(bfloat %a) #0
|
|
declare bfloat @llvm.log2.f16(bfloat %a) #0
|
|
declare bfloat @llvm.fma.f16(bfloat %a, bfloat %b, bfloat %c) #0
|
|
declare bfloat @llvm.fabs.f16(bfloat %a) #0
|
|
declare bfloat @llvm.minnum.f16(bfloat %a, bfloat %b) #0
|
|
declare bfloat @llvm.maxnum.f16(bfloat %a, bfloat %b) #0
|
|
declare bfloat @llvm.copysign.f16(bfloat %a, bfloat %b) #0
|
|
declare bfloat @llvm.floor.f16(bfloat %a) #0
|
|
declare bfloat @llvm.ceil.f16(bfloat %a) #0
|
|
declare bfloat @llvm.trunc.f16(bfloat %a) #0
|
|
declare bfloat @llvm.rint.f16(bfloat %a) #0
|
|
declare bfloat @llvm.nearbyint.f16(bfloat %a) #0
|
|
declare bfloat @llvm.round.f16(bfloat %a) #0
|
|
declare bfloat @llvm.roundeven.f16(bfloat %a) #0
|
|
declare bfloat @llvm.fmuladd.f16(bfloat %a, bfloat %b, bfloat %c) #0
|
|
|
|
|
|
define bfloat @test_sqrt(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_sqrt:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fsqrt s0, s0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_sqrt:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fsqrt s0, s0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.sqrt.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_powi(bfloat %a, i32 %b) #0 {
|
|
; CHECK-CVT-LABEL: test_powi:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl __powisf2
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_powi:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl __powisf2
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.powi.f16.i32(bfloat %a, i32 %b)
|
|
ret bfloat %r
|
|
}
|
|
|
|
|
|
define bfloat @test_sin(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_sin:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl sinf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_sin:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl sinf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.sin.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_cos(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_cos:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl cosf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_cos:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl cosf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.cos.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_tan(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_tan:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl tanf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_tan:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl tanf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.tan.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_acos(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_acos:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl acosf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_acos:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl acosf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.acos.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_asin(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_asin:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl asinf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_asin:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl asinf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.asin.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_atan(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_atan:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl atanf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_atan:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl atanf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.atan.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_atan2(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_atan2:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: // kill: def $s1 killed $s1 killed $q1
|
|
; CHECK-CVT-NEXT: bl atan2f
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_atan2:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: // kill: def $s1 killed $s1 killed $q1
|
|
; CHECK-BF16-NEXT: bl atan2f
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.atan2.f16(bfloat %a, bfloat %b)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_cosh(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_cosh:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl coshf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_cosh:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl coshf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.cosh.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_sinh(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_sinh:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl sinhf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_sinh:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl sinhf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.sinh.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_tanh(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_tanh:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl tanhf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_tanh:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl tanhf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.tanh.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_pow(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_pow:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: // kill: def $s1 killed $s1 killed $q1
|
|
; CHECK-CVT-NEXT: bl powf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_pow:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: // kill: def $s1 killed $s1 killed $q1
|
|
; CHECK-BF16-NEXT: bl powf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.pow.f16(bfloat %a, bfloat %b)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_exp(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_exp:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl expf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_exp:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl expf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.exp.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_exp2(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_exp2:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl exp2f
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_exp2:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl exp2f
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.exp2.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_log(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_log:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl logf
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_log:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl logf
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.log.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_log10(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_log10:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl log10f
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_log10:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl log10f
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.log10.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_log2(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_log2:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: bl log2f
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_log2:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-BF16-NEXT: bl log2f
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.log2.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fma(bfloat %a, bfloat %b, bfloat %c) #0 {
|
|
; CHECK-CVT-LABEL: test_fma:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w10, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v2.4s, v2.4h, #16
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fmadd s0, s0, s1, s2
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: ubfx w9, w8, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w8, w10
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fma:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v2.4s, v2.4h, #16
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fmadd s0, s0, s1, s2
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.fma.f16(bfloat %a, bfloat %b, bfloat %c)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fabs(bfloat %a) #0 {
|
|
; CHECK-LABEL: test_fabs:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 def $s0
|
|
; CHECK-NEXT: fmov w8, s0
|
|
; CHECK-NEXT: and w8, w8, #0x7fff
|
|
; CHECK-NEXT: fmov s0, w8
|
|
; CHECK-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-NEXT: ret
|
|
%r = call bfloat @llvm.fabs.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_minnum(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_minnum:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fminnm s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_minnum:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fminnm s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.minnum.f16(bfloat %a, bfloat %b)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_maxnum(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_maxnum:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fmaxnm s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_maxnum:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fmaxnm s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.maxnum.f16(bfloat %a, bfloat %b)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_copysign(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_copysign:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mvni v2.4s, #128, lsl #24
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-SD-LABEL: test_copysign:
|
|
; CHECK-SD: // %bb.0:
|
|
; CHECK-SD-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-SD-NEXT: mvni v2.4s, #128, lsl #24
|
|
; CHECK-SD-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-SD-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-SD-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-SD-NEXT: bfcvt h0, s0
|
|
; CHECK-SD-NEXT: ret
|
|
;
|
|
; CHECK-GI-LABEL: test_copysign:
|
|
; CHECK-GI: // %bb.0:
|
|
; CHECK-GI-NEXT: mvni v2.8h, #128, lsl #8
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 def $q0
|
|
; CHECK-GI-NEXT: // kill: def $h1 killed $h1 def $q1
|
|
; CHECK-GI-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 killed $q0
|
|
; CHECK-GI-NEXT: ret
|
|
%r = call bfloat @llvm.copysign.f16(bfloat %a, bfloat %b)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_copysign_f32(bfloat %a, float %b) #0 {
|
|
; CHECK-CVT-LABEL: test_copysign_f32:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mvni v2.4s, #128, lsl #24
|
|
; CHECK-CVT-NEXT: // kill: def $s1 killed $s1 def $q1
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-SD-LABEL: test_copysign_f32:
|
|
; CHECK-SD: // %bb.0:
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-SD-NEXT: mvni v2.4s, #128, lsl #24
|
|
; CHECK-SD-NEXT: // kill: def $s1 killed $s1 def $q1
|
|
; CHECK-SD-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-SD-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-SD-NEXT: bfcvt h0, s0
|
|
; CHECK-SD-NEXT: ret
|
|
;
|
|
; CHECK-GI-LABEL: test_copysign_f32:
|
|
; CHECK-GI: // %bb.0:
|
|
; CHECK-GI-NEXT: bfcvt h1, s1
|
|
; CHECK-GI-NEXT: mvni v2.8h, #128, lsl #8
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 def $q0
|
|
; CHECK-GI-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 killed $q0
|
|
; CHECK-GI-NEXT: ret
|
|
%tb = fptrunc float %b to bfloat
|
|
%r = call bfloat @llvm.copysign.f16(bfloat %a, bfloat %tb)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_copysign_f64(bfloat %a, double %b) #0 {
|
|
; CHECK-CVT-LABEL: test_copysign_f64:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: fcvt s1, d1
|
|
; CHECK-CVT-NEXT: mvni v2.4s, #128, lsl #24
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-SD-LABEL: test_copysign_f64:
|
|
; CHECK-SD: // %bb.0:
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-SD-NEXT: fcvt s1, d1
|
|
; CHECK-SD-NEXT: mvni v2.4s, #128, lsl #24
|
|
; CHECK-SD-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-SD-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-SD-NEXT: bfcvt h0, s0
|
|
; CHECK-SD-NEXT: ret
|
|
;
|
|
; CHECK-GI-LABEL: test_copysign_f64:
|
|
; CHECK-GI: // %bb.0:
|
|
; CHECK-GI-NEXT: fcvtxn s1, d1
|
|
; CHECK-GI-NEXT: mvni v2.8h, #128, lsl #8
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 def $q0
|
|
; CHECK-GI-NEXT: bfcvt h1, s1
|
|
; CHECK-GI-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 killed $q0
|
|
; CHECK-GI-NEXT: ret
|
|
%tb = fptrunc double %b to bfloat
|
|
%r = call bfloat @llvm.copysign.f16(bfloat %a, bfloat %tb)
|
|
ret bfloat %r
|
|
}
|
|
|
|
; away the (fpext (fp_round <result>)) here.
|
|
|
|
define float @test_copysign_extended(bfloat %a, bfloat %b) #0 {
|
|
; CHECK-CVT-LABEL: test_copysign_extended:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mvni v2.4s, #128, lsl #24
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-SD-LABEL: test_copysign_extended:
|
|
; CHECK-SD: // %bb.0:
|
|
; CHECK-SD-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-SD-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-SD-NEXT: mvni v2.4s, #128, lsl #24
|
|
; CHECK-SD-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-SD-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-SD-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-SD-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-SD-NEXT: ret
|
|
;
|
|
; CHECK-GI-LABEL: test_copysign_extended:
|
|
; CHECK-GI: // %bb.0:
|
|
; CHECK-GI-NEXT: mvni v2.8h, #128, lsl #8
|
|
; CHECK-GI-NEXT: // kill: def $h0 killed $h0 def $q0
|
|
; CHECK-GI-NEXT: // kill: def $h1 killed $h1 def $q1
|
|
; CHECK-GI-NEXT: bif v0.16b, v1.16b, v2.16b
|
|
; CHECK-GI-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-GI-NEXT: // kill: def $s0 killed $s0 killed $q0
|
|
; CHECK-GI-NEXT: ret
|
|
%r = call bfloat @llvm.copysign.f16(bfloat %a, bfloat %b)
|
|
%xr = fpext bfloat %r to float
|
|
ret float %xr
|
|
}
|
|
|
|
define bfloat @test_floor(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_floor:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: frintm s0, s0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_floor:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: frintm s0, s0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.floor.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_ceil(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_ceil:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: frintp s0, s0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_ceil:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: frintp s0, s0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.ceil.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_trunc(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_trunc:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: frintz s0, s0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_trunc:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: frintz s0, s0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.trunc.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_rint(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_rint:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: frintx s0, s0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_rint:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: frintx s0, s0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.rint.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_nearbyint(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_nearbyint:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: frinti s0, s0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_nearbyint:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: frinti s0, s0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.nearbyint.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_round(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_round:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: frinta s0, s0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_round:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: frinta s0, s0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.round.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_roundeven(bfloat %a) #0 {
|
|
; CHECK-CVT-LABEL: test_roundeven:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w8, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: frintn s0, s0
|
|
; CHECK-CVT-NEXT: fmov w9, s0
|
|
; CHECK-CVT-NEXT: ubfx w10, w9, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: add w8, w10, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_roundeven:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: frintn s0, s0
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.roundeven.f16(bfloat %a)
|
|
ret bfloat %r
|
|
}
|
|
|
|
define bfloat @test_fmuladd(bfloat %a, bfloat %b, bfloat %c) #0 {
|
|
; CHECK-CVT-LABEL: test_fmuladd:
|
|
; CHECK-CVT: // %bb.0:
|
|
; CHECK-CVT-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-CVT-NEXT: mov w10, #32767 // =0x7fff
|
|
; CHECK-CVT-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-CVT-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fmul s0, s0, s1
|
|
; CHECK-CVT-NEXT: shll v1.4s, v2.4h, #16
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: ubfx w9, w8, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w8, w10
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-CVT-NEXT: fadd s0, s0, s1
|
|
; CHECK-CVT-NEXT: fmov w8, s0
|
|
; CHECK-CVT-NEXT: ubfx w9, w8, #16, #1
|
|
; CHECK-CVT-NEXT: add w8, w8, w10
|
|
; CHECK-CVT-NEXT: add w8, w9, w8
|
|
; CHECK-CVT-NEXT: lsr w8, w8, #16
|
|
; CHECK-CVT-NEXT: fmov s0, w8
|
|
; CHECK-CVT-NEXT: // kill: def $h0 killed $h0 killed $s0
|
|
; CHECK-CVT-NEXT: ret
|
|
;
|
|
; CHECK-BF16-LABEL: test_fmuladd:
|
|
; CHECK-BF16: // %bb.0:
|
|
; CHECK-BF16-NEXT: // kill: def $h1 killed $h1 def $d1
|
|
; CHECK-BF16-NEXT: // kill: def $h0 killed $h0 def $d0
|
|
; CHECK-BF16-NEXT: // kill: def $h2 killed $h2 def $d2
|
|
; CHECK-BF16-NEXT: shll v1.4s, v1.4h, #16
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fmul s0, s0, s1
|
|
; CHECK-BF16-NEXT: shll v1.4s, v2.4h, #16
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: shll v0.4s, v0.4h, #16
|
|
; CHECK-BF16-NEXT: fadd s0, s0, s1
|
|
; CHECK-BF16-NEXT: bfcvt h0, s0
|
|
; CHECK-BF16-NEXT: ret
|
|
%r = call bfloat @llvm.fmuladd.f16(bfloat %a, bfloat %b, bfloat %c)
|
|
ret bfloat %r
|
|
}
|
|
|
|
attributes #0 = { nounwind }
|