Home | History | Annotate | Download | only in SystemZ
      1 ; Testg 64-bit signed division and remainder.
      2 ;
      3 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
      4 
      5 declare i64 @foo()
      6 
      7 ; Testg register division.  The result is in the second of the two registers.
      8 define void @f1(i64 %dummy, i64 %a, i64 %b, i64 *%dest) {
      9 ; CHECK-LABEL: f1:
     10 ; CHECK-NOT: {{%r[234]}}
     11 ; CHECK: dsgr %r2, %r4
     12 ; CHECK: stg %r3, 0(%r5)
     13 ; CHECK: br %r14
     14   %div = sdiv i64 %a, %b
     15   store i64 %div, i64 *%dest
     16   ret void
     17 }
     18 
     19 ; Testg register remainder.  The result is in the first of the two registers.
     20 define void @f2(i64 %dummy, i64 %a, i64 %b, i64 *%dest) {
     21 ; CHECK-LABEL: f2:
     22 ; CHECK-NOT: {{%r[234]}}
     23 ; CHECK: dsgr %r2, %r4
     24 ; CHECK: stg %r2, 0(%r5)
     25 ; CHECK: br %r14
     26   %rem = srem i64 %a, %b
     27   store i64 %rem, i64 *%dest
     28   ret void
     29 }
     30 
     31 ; Testg that division and remainder use a single instruction.
     32 define i64 @f3(i64 %dummy1, i64 %a, i64 %b) {
     33 ; CHECK-LABEL: f3:
     34 ; CHECK-NOT: {{%r[234]}}
     35 ; CHECK: dsgr %r2, %r4
     36 ; CHECK-NOT: dsgr
     37 ; CHECK: ogr %r2, %r3
     38 ; CHECK: br %r14
     39   %div = sdiv i64 %a, %b
     40   %rem = srem i64 %a, %b
     41   %or = or i64 %rem, %div
     42   ret i64 %or
     43 }
     44 
     45 ; Testg memory division with no displacement.
     46 define void @f4(i64 %dummy, i64 %a, i64 *%src, i64 *%dest) {
     47 ; CHECK-LABEL: f4:
     48 ; CHECK-NOT: {{%r[234]}}
     49 ; CHECK: dsg %r2, 0(%r4)
     50 ; CHECK: stg %r3, 0(%r5)
     51 ; CHECK: br %r14
     52   %b = load i64 *%src
     53   %div = sdiv i64 %a, %b
     54   store i64 %div, i64 *%dest
     55   ret void
     56 }
     57 
     58 ; Testg memory remainder with no displacement.
     59 define void @f5(i64 %dummy, i64 %a, i64 *%src, i64 *%dest) {
     60 ; CHECK-LABEL: f5:
     61 ; CHECK-NOT: {{%r[234]}}
     62 ; CHECK: dsg %r2, 0(%r4)
     63 ; CHECK: stg %r2, 0(%r5)
     64 ; CHECK: br %r14
     65   %b = load i64 *%src
     66   %rem = srem i64 %a, %b
     67   store i64 %rem, i64 *%dest
     68   ret void
     69 }
     70 
     71 ; Testg both memory division and memory remainder.
     72 define i64 @f6(i64 %dummy, i64 %a, i64 *%src) {
     73 ; CHECK-LABEL: f6:
     74 ; CHECK-NOT: {{%r[234]}}
     75 ; CHECK: dsg %r2, 0(%r4)
     76 ; CHECK-NOT: {{dsg|dsgr}}
     77 ; CHECK: ogr %r2, %r3
     78 ; CHECK: br %r14
     79   %b = load i64 *%src
     80   %div = sdiv i64 %a, %b
     81   %rem = srem i64 %a, %b
     82   %or = or i64 %rem, %div
     83   ret i64 %or
     84 }
     85 
     86 ; Check the high end of the DSG range.
     87 define i64 @f7(i64 %dummy, i64 %a, i64 *%src) {
     88 ; CHECK-LABEL: f7:
     89 ; CHECK: dsg %r2, 524280(%r4)
     90 ; CHECK: br %r14
     91   %ptr = getelementptr i64 *%src, i64 65535
     92   %b = load i64 *%ptr
     93   %rem = srem i64 %a, %b
     94   ret i64 %rem
     95 }
     96 
     97 ; Check the next doubleword up, which needs separate address logic.
     98 ; Other sequences besides this one would be OK.
     99 define i64 @f8(i64 %dummy, i64 %a, i64 *%src) {
    100 ; CHECK-LABEL: f8:
    101 ; CHECK: agfi %r4, 524288
    102 ; CHECK: dsg %r2, 0(%r4)
    103 ; CHECK: br %r14
    104   %ptr = getelementptr i64 *%src, i64 65536
    105   %b = load i64 *%ptr
    106   %rem = srem i64 %a, %b
    107   ret i64 %rem
    108 }
    109 
    110 ; Check the high end of the negative aligned DSG range.
    111 define i64 @f9(i64 %dummy, i64 %a, i64 *%src) {
    112 ; CHECK-LABEL: f9:
    113 ; CHECK: dsg %r2, -8(%r4)
    114 ; CHECK: br %r14
    115   %ptr = getelementptr i64 *%src, i64 -1
    116   %b = load i64 *%ptr
    117   %rem = srem i64 %a, %b
    118   ret i64 %rem
    119 }
    120 
    121 ; Check the low end of the DSG range.
    122 define i64 @f10(i64 %dummy, i64 %a, i64 *%src) {
    123 ; CHECK-LABEL: f10:
    124 ; CHECK: dsg %r2, -524288(%r4)
    125 ; CHECK: br %r14
    126   %ptr = getelementptr i64 *%src, i64 -65536
    127   %b = load i64 *%ptr
    128   %rem = srem i64 %a, %b
    129   ret i64 %rem
    130 }
    131 
    132 ; Check the next doubleword down, which needs separate address logic.
    133 ; Other sequences besides this one would be OK.
    134 define i64 @f11(i64 %dummy, i64 %a, i64 *%src) {
    135 ; CHECK-LABEL: f11:
    136 ; CHECK: agfi %r4, -524296
    137 ; CHECK: dsg %r2, 0(%r4)
    138 ; CHECK: br %r14
    139   %ptr = getelementptr i64 *%src, i64 -65537
    140   %b = load i64 *%ptr
    141   %rem = srem i64 %a, %b
    142   ret i64 %rem
    143 }
    144 
    145 ; Check that DSG allows an index.
    146 define i64 @f12(i64 %dummy, i64 %a, i64 %src, i64 %index) {
    147 ; CHECK-LABEL: f12:
    148 ; CHECK: dsg %r2, 524287(%r5,%r4)
    149 ; CHECK: br %r14
    150   %add1 = add i64 %src, %index
    151   %add2 = add i64 %add1, 524287
    152   %ptr = inttoptr i64 %add2 to i64 *
    153   %b = load i64 *%ptr
    154   %rem = srem i64 %a, %b
    155   ret i64 %rem
    156 }
    157 
    158 ; Check that divisions of spilled values can use DSG rather than DSGR.
    159 define i64 @f13(i64 *%ptr0) {
    160 ; CHECK-LABEL: f13:
    161 ; CHECK: brasl %r14, foo@PLT
    162 ; CHECK: dsg {{%r[0-9]+}}, 160(%r15)
    163 ; CHECK: br %r14
    164   %ptr1 = getelementptr i64 *%ptr0, i64 2
    165   %ptr2 = getelementptr i64 *%ptr0, i64 4
    166   %ptr3 = getelementptr i64 *%ptr0, i64 6
    167   %ptr4 = getelementptr i64 *%ptr0, i64 8
    168   %ptr5 = getelementptr i64 *%ptr0, i64 10
    169   %ptr6 = getelementptr i64 *%ptr0, i64 12
    170   %ptr7 = getelementptr i64 *%ptr0, i64 14
    171   %ptr8 = getelementptr i64 *%ptr0, i64 16
    172   %ptr9 = getelementptr i64 *%ptr0, i64 18
    173   %ptr10 = getelementptr i64 *%ptr0, i64 20
    174 
    175   %val0 = load i64 *%ptr0
    176   %val1 = load i64 *%ptr1
    177   %val2 = load i64 *%ptr2
    178   %val3 = load i64 *%ptr3
    179   %val4 = load i64 *%ptr4
    180   %val5 = load i64 *%ptr5
    181   %val6 = load i64 *%ptr6
    182   %val7 = load i64 *%ptr7
    183   %val8 = load i64 *%ptr8
    184   %val9 = load i64 *%ptr9
    185   %val10 = load i64 *%ptr10
    186 
    187   %ret = call i64 @foo()
    188 
    189   %div0 = sdiv i64 %ret, %val0
    190   %div1 = sdiv i64 %div0, %val1
    191   %div2 = sdiv i64 %div1, %val2
    192   %div3 = sdiv i64 %div2, %val3
    193   %div4 = sdiv i64 %div3, %val4
    194   %div5 = sdiv i64 %div4, %val5
    195   %div6 = sdiv i64 %div5, %val6
    196   %div7 = sdiv i64 %div6, %val7
    197   %div8 = sdiv i64 %div7, %val8
    198   %div9 = sdiv i64 %div8, %val9
    199   %div10 = sdiv i64 %div9, %val10
    200 
    201   ret i64 %div10
    202 }
    203