Home | History | Annotate | Download | only in Thumb2
      1 ; RUN: llc < %s -mtriple=thumbv7-none-eabi   -mcpu=cortex-m3                    | FileCheck %s -check-prefix=CHECK -check-prefix=SOFT -check-prefix=NONE
      2 ; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-m4                    | FileCheck %s -check-prefix=CHECK -check-prefix=SOFT -check-prefix=SP
      3 ; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-m7                    | FileCheck %s -check-prefix=CHECK -check-prefix=HARD -check-prefix=DP -check-prefix=VFP  -check-prefix=FP-ARMv8
      4 ; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-m7 -mattr=+fp-only-sp | FileCheck %s -check-prefix=CHECK -check-prefix=SOFT -check-prefix=SP
      5 ; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-a7                    | FileCheck %s -check-prefix=CHECK -check-prefix=HARD -check-prefix=DP -check-prefix=NEON -check-prefix=VFP4
      6 ; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-a57                   | FileCheck %s -check-prefix=CHECK -check-prefix=HARD -check-prefix=DP -check-prefix=NEON -check-prefix=FP-ARMv8
      7 
      8 declare double     @llvm.sqrt.f64(double %Val)
      9 define double @sqrt_d(double %a) {
     10 ; CHECK-LABEL: sqrt_d:
     11 ; SOFT: {{(bl|b)}} sqrt
     12 ; HARD: vsqrt.f64 d0, d0
     13   %1 = call double @llvm.sqrt.f64(double %a)
     14   ret double %1
     15 }
     16 
     17 declare double     @llvm.powi.f64(double %Val, i32 %power)
     18 define double @powi_d(double %a, i32 %b) {
     19 ; CHECK-LABEL: powi_d:
     20 ; SOFT: {{(bl|b)}} __powidf2
     21 ; HARD: b __powidf2
     22   %1 = call double @llvm.powi.f64(double %a, i32 %b)
     23   ret double %1
     24 }
     25 
     26 declare double     @llvm.sin.f64(double %Val)
     27 define double @sin_d(double %a) {
     28 ; CHECK-LABEL: sin_d:
     29 ; SOFT: {{(bl|b)}} sin
     30 ; HARD: b sin
     31   %1 = call double @llvm.sin.f64(double %a)
     32   ret double %1
     33 }
     34 
     35 declare double     @llvm.cos.f64(double %Val)
     36 define double @cos_d(double %a) {
     37 ; CHECK-LABEL: cos_d:
     38 ; SOFT: {{(bl|b)}} cos
     39 ; HARD: b cos
     40   %1 = call double @llvm.cos.f64(double %a)
     41   ret double %1
     42 }
     43 
     44 declare double     @llvm.pow.f64(double %Val, double %power)
     45 define double @pow_d(double %a, double %b) {
     46 ; CHECK-LABEL: pow_d:
     47 ; SOFT: {{(bl|b)}} pow
     48 ; HARD: b pow
     49   %1 = call double @llvm.pow.f64(double %a, double %b)
     50   ret double %1
     51 }
     52 
     53 declare double     @llvm.exp.f64(double %Val)
     54 define double @exp_d(double %a) {
     55 ; CHECK-LABEL: exp_d:
     56 ; SOFT: {{(bl|b)}} exp
     57 ; HARD: b exp
     58   %1 = call double @llvm.exp.f64(double %a)
     59   ret double %1
     60 }
     61 
     62 declare double     @llvm.exp2.f64(double %Val)
     63 define double @exp2_d(double %a) {
     64 ; CHECK-LABEL: exp2_d:
     65 ; SOFT: {{(bl|b)}} exp2
     66 ; HARD: b exp2
     67   %1 = call double @llvm.exp2.f64(double %a)
     68   ret double %1
     69 }
     70 
     71 declare double     @llvm.log.f64(double %Val)
     72 define double @log_d(double %a) {
     73 ; CHECK-LABEL: log_d:
     74 ; SOFT: {{(bl|b)}} log
     75 ; HARD: b log
     76   %1 = call double @llvm.log.f64(double %a)
     77   ret double %1
     78 }
     79 
     80 declare double     @llvm.log10.f64(double %Val)
     81 define double @log10_d(double %a) {
     82 ; CHECK-LABEL: log10_d:
     83 ; SOFT: {{(bl|b)}} log10
     84 ; HARD: b log10
     85   %1 = call double @llvm.log10.f64(double %a)
     86   ret double %1
     87 }
     88 
     89 declare double     @llvm.log2.f64(double %Val)
     90 define double @log2_d(double %a) {
     91 ; CHECK-LABEL: log2_d:
     92 ; SOFT: {{(bl|b)}} log2
     93 ; HARD: b log2
     94   %1 = call double @llvm.log2.f64(double %a)
     95   ret double %1
     96 }
     97 
     98 declare double     @llvm.fma.f64(double %a, double %b, double %c)
     99 define double @fma_d(double %a, double %b, double %c) {
    100 ; CHECK-LABEL: fma_d:
    101 ; SOFT: {{(bl|b)}} fma
    102 ; HARD: vfma.f64
    103   %1 = call double @llvm.fma.f64(double %a, double %b, double %c)
    104   ret double %1
    105 }
    106 
    107 ; FIXME: the FPv4-SP version is less efficient than the no-FPU version
    108 declare double     @llvm.fabs.f64(double %Val)
    109 define double @abs_d(double %a) {
    110 ; CHECK-LABEL: abs_d:
    111 ; NONE: bic r1, r1, #-2147483648
    112 ; SP: bl __aeabi_dcmpgt
    113 ; SP: bl __aeabi_dcmpun
    114 ; SP: bl __aeabi_dsub
    115 ; DP: vabs.f64 d0, d0
    116   %1 = call double @llvm.fabs.f64(double %a)
    117   ret double %1
    118 }
    119 
    120 declare double     @llvm.copysign.f64(double  %Mag, double  %Sgn)
    121 define double @copysign_d(double %a, double %b) {
    122 ; CHECK-LABEL: copysign_d:
    123 ; SOFT: lsrs [[REG:r[0-9]+]], r3, #31
    124 ; SOFT: bfi r1, [[REG]], #31, #1
    125 ; VFP: lsrs [[REG:r[0-9]+]], r3, #31
    126 ; VFP: bfi r1, [[REG]], #31, #1
    127 ; NEON: vmov.i32 [[REG:d[0-9]+]], #0x80000000
    128 ; NEON: vshl.i64 [[REG]], [[REG]], #32
    129 ; NEON: vbsl [[REG]], d
    130   %1 = call double @llvm.copysign.f64(double %a, double %b)
    131   ret double %1
    132 }
    133 
    134 declare double     @llvm.floor.f64(double %Val)
    135 define double @floor_d(double %a) {
    136 ; CHECK-LABEL: floor_d:
    137 ; SOFT: {{(bl|b)}} floor
    138 ; VFP4: b floor
    139 ; FP-ARMv8: vrintm.f64
    140   %1 = call double @llvm.floor.f64(double %a)
    141   ret double %1
    142 }
    143 
    144 declare double     @llvm.ceil.f64(double %Val)
    145 define double @ceil_d(double %a) {
    146 ; CHECK-LABEL: ceil_d:
    147 ; SOFT: {{(bl|b)}} ceil
    148 ; VFP4: b ceil
    149 ; FP-ARMv8: vrintp.f64
    150   %1 = call double @llvm.ceil.f64(double %a)
    151   ret double %1
    152 }
    153 
    154 declare double     @llvm.trunc.f64(double %Val)
    155 define double @trunc_d(double %a) {
    156 ; CHECK-LABEL: trunc_d:
    157 ; SOFT: {{(bl|b)}} trunc
    158 ; FFP4: b trunc
    159 ; FP-ARMv8: vrintz.f64
    160   %1 = call double @llvm.trunc.f64(double %a)
    161   ret double %1
    162 }
    163 
    164 declare double     @llvm.rint.f64(double %Val)
    165 define double @rint_d(double %a) {
    166 ; CHECK-LABEL: rint_d:
    167 ; SOFT: {{(bl|b)}} rint
    168 ; VFP4: b rint
    169 ; FP-ARMv8: vrintx.f64
    170   %1 = call double @llvm.rint.f64(double %a)
    171   ret double %1
    172 }
    173 
    174 declare double     @llvm.nearbyint.f64(double %Val)
    175 define double @nearbyint_d(double %a) {
    176 ; CHECK-LABEL: nearbyint_d:
    177 ; SOFT: {{(bl|b)}} nearbyint
    178 ; VFP4: b nearbyint
    179 ; FP-ARMv8: vrintr.f64
    180   %1 = call double @llvm.nearbyint.f64(double %a)
    181   ret double %1
    182 }
    183 
    184 declare double     @llvm.round.f64(double %Val)
    185 define double @round_d(double %a) {
    186 ; CHECK-LABEL: round_d:
    187 ; SOFT: {{(bl|b)}} round
    188 ; VFP4: b round
    189 ; FP-ARMv8: vrinta.f64
    190   %1 = call double @llvm.round.f64(double %a)
    191   ret double %1
    192 }
    193 
    194 declare double     @llvm.fmuladd.f64(double %a, double %b, double %c)
    195 define double @fmuladd_d(double %a, double %b, double %c) {
    196 ; CHECK-LABEL: fmuladd_d:
    197 ; SOFT: bl __aeabi_dmul
    198 ; SOFT: bl __aeabi_dadd
    199 ; VFP4: vmul.f64
    200 ; VFP4: vadd.f64
    201 ; FP-ARMv8: vmla.f64
    202   %1 = call double @llvm.fmuladd.f64(double %a, double %b, double %c)
    203   ret double %1
    204 }
    205 
    206 declare i16 @llvm.convert.to.fp16.f64(double %a)
    207 define i16 @d_to_h(double %a) {
    208 ; CHECK-LABEL: d_to_h:
    209 ; SOFT: bl __aeabi_d2h
    210 ; VFP4: bl __aeabi_d2h
    211 ; FP-ARMv8: vcvt{{[bt]}}.f16.f64
    212   %1 = call i16 @llvm.convert.to.fp16.f64(double %a)
    213   ret i16 %1
    214 }
    215 
    216 declare double @llvm.convert.from.fp16.f64(i16 %a)
    217 define double @h_to_d(i16 %a) {
    218 ; CHECK-LABEL: h_to_d:
    219 ; NONE: bl __gnu_h2f_ieee
    220 ; NONE: bl __aeabi_f2d
    221 ; SP: vcvt{{[bt]}}.f32.f16
    222 ; SP: bl __aeabi_f2d
    223 ; VFPv4: vcvt{{[bt]}}.f32.f16
    224 ; VFPv4: vcvt.f64.f32
    225 ; FP-ARMv8: vcvt{{[bt]}}.f64.f16
    226   %1 = call double @llvm.convert.from.fp16.f64(i16 %a)
    227   ret double %1
    228 }
    229