1 ; Test the saving and restoring of GPRs in large frames. 2 ; 3 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s 4 5 ; This is the largest frame size that can use a plain LMG for %r6 and above. 6 ; It is big enough to require two emergency spill slots at 160(%r15), 7 ; so get a frame of size 524232 by allocating (524232 - 176) / 8 = 65507 8 ; extra doublewords. 9 define void @f1(i32 *%ptr, i64 %x) { 10 ; CHECK-LABEL: f1: 11 ; CHECK: stmg %r6, %r15, 48(%r15) 12 ; CHECK: .cfi_offset %r6, -112 13 ; CHECK: .cfi_offset %r7, -104 14 ; CHECK: .cfi_offset %r8, -96 15 ; CHECK: .cfi_offset %r9, -88 16 ; CHECK: .cfi_offset %r10, -80 17 ; CHECK: .cfi_offset %r11, -72 18 ; CHECK: .cfi_offset %r12, -64 19 ; CHECK: .cfi_offset %r13, -56 20 ; CHECK: .cfi_offset %r14, -48 21 ; CHECK: .cfi_offset %r15, -40 22 ; CHECK: agfi %r15, -524232 23 ; CHECK: .cfi_def_cfa_offset 524392 24 ; ...main function body... 25 ; CHECK-NOT: ag 26 ; CHECK: lmg %r6, %r15, 524280(%r15) 27 ; CHECK: br %r14 28 %l0 = load volatile i32 *%ptr 29 %l1 = load volatile i32 *%ptr 30 %l4 = load volatile i32 *%ptr 31 %l5 = load volatile i32 *%ptr 32 %l6 = load volatile i32 *%ptr 33 %l7 = load volatile i32 *%ptr 34 %l8 = load volatile i32 *%ptr 35 %l9 = load volatile i32 *%ptr 36 %l10 = load volatile i32 *%ptr 37 %l11 = load volatile i32 *%ptr 38 %l12 = load volatile i32 *%ptr 39 %l13 = load volatile i32 *%ptr 40 %l14 = load volatile i32 *%ptr 41 %add0 = add i32 %l0, %l0 42 %add1 = add i32 %l1, %add0 43 %add4 = add i32 %l4, %add1 44 %add5 = add i32 %l5, %add4 45 %add6 = add i32 %l6, %add5 46 %add7 = add i32 %l7, %add6 47 %add8 = add i32 %l8, %add7 48 %add9 = add i32 %l9, %add8 49 %add10 = add i32 %l10, %add9 50 %add11 = add i32 %l11, %add10 51 %add12 = add i32 %l12, %add11 52 %add13 = add i32 %l13, %add12 53 %add14 = add i32 %l14, %add13 54 store volatile i32 %add0, i32 *%ptr 55 store volatile i32 %add1, i32 *%ptr 56 store volatile i32 %add4, i32 *%ptr 57 store volatile i32 %add5, i32 *%ptr 58 store volatile i32 %add6, i32 *%ptr 59 store volatile i32 %add7, i32 *%ptr 60 store volatile i32 %add8, i32 *%ptr 61 store volatile i32 %add9, i32 *%ptr 62 store volatile i32 %add10, i32 *%ptr 63 store volatile i32 %add11, i32 *%ptr 64 store volatile i32 %add12, i32 *%ptr 65 store volatile i32 %add13, i32 *%ptr 66 store volatile i32 %add14, i32 *%ptr 67 %y = alloca [65507 x i64], align 8 68 %entry = getelementptr inbounds [65507 x i64]* %y, i64 0, i64 0 69 store volatile i64 %x, i64* %entry 70 ret void 71 } 72 73 ; This is the largest frame size that can use a plain LMG for %r14 and above 74 ; It is big enough to require two emergency spill slots at 160(%r15), 75 ; so get a frame of size 524168 by allocating (524168 - 176) / 8 = 65499 76 ; extra doublewords. 77 define void @f2(i32 *%ptr, i64 %x) { 78 ; CHECK-LABEL: f2: 79 ; CHECK: stmg %r14, %r15, 112(%r15) 80 ; CHECK: .cfi_offset %r14, -48 81 ; CHECK: .cfi_offset %r15, -40 82 ; CHECK: agfi %r15, -524168 83 ; CHECK: .cfi_def_cfa_offset 524328 84 ; ...main function body... 85 ; CHECK-NOT: ag 86 ; CHECK: lmg %r14, %r15, 524280(%r15) 87 ; CHECK: br %r14 88 %l0 = load volatile i32 *%ptr 89 %l1 = load volatile i32 *%ptr 90 %l4 = load volatile i32 *%ptr 91 %l5 = load volatile i32 *%ptr 92 %l14 = load volatile i32 *%ptr 93 %add0 = add i32 %l0, %l0 94 %add1 = add i32 %l1, %add0 95 %add4 = add i32 %l4, %add1 96 %add5 = add i32 %l5, %add4 97 %add14 = add i32 %l14, %add5 98 store volatile i32 %add0, i32 *%ptr 99 store volatile i32 %add1, i32 *%ptr 100 store volatile i32 %add4, i32 *%ptr 101 store volatile i32 %add5, i32 *%ptr 102 store volatile i32 %add14, i32 *%ptr 103 %y = alloca [65499 x i64], align 8 104 %entry = getelementptr inbounds [65499 x i64]* %y, i64 0, i64 0 105 store volatile i64 %x, i64* %entry 106 ret void 107 } 108 109 ; Like f1 but with a frame that is 8 bytes bigger. This is the smallest 110 ; frame size that needs two instructions to perform the final LMG for 111 ; %r6 and above. 112 define void @f3(i32 *%ptr, i64 %x) { 113 ; CHECK-LABEL: f3: 114 ; CHECK: stmg %r6, %r15, 48(%r15) 115 ; CHECK: .cfi_offset %r6, -112 116 ; CHECK: .cfi_offset %r7, -104 117 ; CHECK: .cfi_offset %r8, -96 118 ; CHECK: .cfi_offset %r9, -88 119 ; CHECK: .cfi_offset %r10, -80 120 ; CHECK: .cfi_offset %r11, -72 121 ; CHECK: .cfi_offset %r12, -64 122 ; CHECK: .cfi_offset %r13, -56 123 ; CHECK: .cfi_offset %r14, -48 124 ; CHECK: .cfi_offset %r15, -40 125 ; CHECK: agfi %r15, -524240 126 ; CHECK: .cfi_def_cfa_offset 524400 127 ; ...main function body... 128 ; CHECK: aghi %r15, 8 129 ; CHECK: lmg %r6, %r15, 524280(%r15) 130 ; CHECK: br %r14 131 %l0 = load volatile i32 *%ptr 132 %l1 = load volatile i32 *%ptr 133 %l4 = load volatile i32 *%ptr 134 %l5 = load volatile i32 *%ptr 135 %l6 = load volatile i32 *%ptr 136 %l7 = load volatile i32 *%ptr 137 %l8 = load volatile i32 *%ptr 138 %l9 = load volatile i32 *%ptr 139 %l10 = load volatile i32 *%ptr 140 %l11 = load volatile i32 *%ptr 141 %l12 = load volatile i32 *%ptr 142 %l13 = load volatile i32 *%ptr 143 %l14 = load volatile i32 *%ptr 144 %add0 = add i32 %l0, %l0 145 %add1 = add i32 %l1, %add0 146 %add4 = add i32 %l4, %add1 147 %add5 = add i32 %l5, %add4 148 %add6 = add i32 %l6, %add5 149 %add7 = add i32 %l7, %add6 150 %add8 = add i32 %l8, %add7 151 %add9 = add i32 %l9, %add8 152 %add10 = add i32 %l10, %add9 153 %add11 = add i32 %l11, %add10 154 %add12 = add i32 %l12, %add11 155 %add13 = add i32 %l13, %add12 156 %add14 = add i32 %l14, %add13 157 store volatile i32 %add0, i32 *%ptr 158 store volatile i32 %add1, i32 *%ptr 159 store volatile i32 %add4, i32 *%ptr 160 store volatile i32 %add5, i32 *%ptr 161 store volatile i32 %add6, i32 *%ptr 162 store volatile i32 %add7, i32 *%ptr 163 store volatile i32 %add8, i32 *%ptr 164 store volatile i32 %add9, i32 *%ptr 165 store volatile i32 %add10, i32 *%ptr 166 store volatile i32 %add11, i32 *%ptr 167 store volatile i32 %add12, i32 *%ptr 168 store volatile i32 %add13, i32 *%ptr 169 store volatile i32 %add14, i32 *%ptr 170 %y = alloca [65508 x i64], align 8 171 %entry = getelementptr inbounds [65508 x i64]* %y, i64 0, i64 0 172 store volatile i64 %x, i64* %entry 173 ret void 174 } 175 176 ; Like f2 but with a frame that is 8 bytes bigger. This is the smallest 177 ; frame size that needs two instructions to perform the final LMG for 178 ; %r14 and %r15. 179 define void @f4(i32 *%ptr, i64 %x) { 180 ; CHECK-LABEL: f4: 181 ; CHECK: stmg %r14, %r15, 112(%r15) 182 ; CHECK: .cfi_offset %r14, -48 183 ; CHECK: .cfi_offset %r15, -40 184 ; CHECK: agfi %r15, -524176 185 ; CHECK: .cfi_def_cfa_offset 524336 186 ; ...main function body... 187 ; CHECK: aghi %r15, 8 188 ; CHECK: lmg %r14, %r15, 524280(%r15) 189 ; CHECK: br %r14 190 %l0 = load volatile i32 *%ptr 191 %l1 = load volatile i32 *%ptr 192 %l4 = load volatile i32 *%ptr 193 %l5 = load volatile i32 *%ptr 194 %l14 = load volatile i32 *%ptr 195 %add0 = add i32 %l0, %l0 196 %add1 = add i32 %l1, %add0 197 %add4 = add i32 %l4, %add1 198 %add5 = add i32 %l5, %add4 199 %add14 = add i32 %l14, %add5 200 store volatile i32 %add0, i32 *%ptr 201 store volatile i32 %add1, i32 *%ptr 202 store volatile i32 %add4, i32 *%ptr 203 store volatile i32 %add5, i32 *%ptr 204 store volatile i32 %add14, i32 *%ptr 205 %y = alloca [65500 x i64], align 8 206 %entry = getelementptr inbounds [65500 x i64]* %y, i64 0, i64 0 207 store volatile i64 %x, i64* %entry 208 ret void 209 } 210 211 ; This is the largest frame size for which the preparatory increment for 212 ; "lmg %r14, %r15, ..." can be done using AGHI. 213 define void @f5(i32 *%ptr, i64 %x) { 214 ; CHECK-LABEL: f5: 215 ; CHECK: stmg %r14, %r15, 112(%r15) 216 ; CHECK: .cfi_offset %r14, -48 217 ; CHECK: .cfi_offset %r15, -40 218 ; CHECK: agfi %r15, -556928 219 ; CHECK: .cfi_def_cfa_offset 557088 220 ; ...main function body... 221 ; CHECK: aghi %r15, 32760 222 ; CHECK: lmg %r14, %r15, 524280(%r15) 223 ; CHECK: br %r14 224 %l0 = load volatile i32 *%ptr 225 %l1 = load volatile i32 *%ptr 226 %l4 = load volatile i32 *%ptr 227 %l5 = load volatile i32 *%ptr 228 %l14 = load volatile i32 *%ptr 229 %add0 = add i32 %l0, %l0 230 %add1 = add i32 %l1, %add0 231 %add4 = add i32 %l4, %add1 232 %add5 = add i32 %l5, %add4 233 %add14 = add i32 %l14, %add5 234 store volatile i32 %add0, i32 *%ptr 235 store volatile i32 %add1, i32 *%ptr 236 store volatile i32 %add4, i32 *%ptr 237 store volatile i32 %add5, i32 *%ptr 238 store volatile i32 %add14, i32 *%ptr 239 %y = alloca [69594 x i64], align 8 240 %entry = getelementptr inbounds [69594 x i64]* %y, i64 0, i64 0 241 store volatile i64 %x, i64* %entry 242 ret void 243 } 244 245 ; This is the smallest frame size for which the preparatory increment for 246 ; "lmg %r14, %r15, ..." needs to be done using AGFI. 247 define void @f6(i32 *%ptr, i64 %x) { 248 ; CHECK-LABEL: f6: 249 ; CHECK: stmg %r14, %r15, 112(%r15) 250 ; CHECK: .cfi_offset %r14, -48 251 ; CHECK: .cfi_offset %r15, -40 252 ; CHECK: agfi %r15, -556936 253 ; CHECK: .cfi_def_cfa_offset 557096 254 ; ...main function body... 255 ; CHECK: agfi %r15, 32768 256 ; CHECK: lmg %r14, %r15, 524280(%r15) 257 ; CHECK: br %r14 258 %l0 = load volatile i32 *%ptr 259 %l1 = load volatile i32 *%ptr 260 %l4 = load volatile i32 *%ptr 261 %l5 = load volatile i32 *%ptr 262 %l14 = load volatile i32 *%ptr 263 %add0 = add i32 %l0, %l0 264 %add1 = add i32 %l1, %add0 265 %add4 = add i32 %l4, %add1 266 %add5 = add i32 %l5, %add4 267 %add14 = add i32 %l14, %add5 268 store volatile i32 %add0, i32 *%ptr 269 store volatile i32 %add1, i32 *%ptr 270 store volatile i32 %add4, i32 *%ptr 271 store volatile i32 %add5, i32 *%ptr 272 store volatile i32 %add14, i32 *%ptr 273 %y = alloca [69595 x i64], align 8 274 %entry = getelementptr inbounds [69595 x i64]* %y, i64 0, i64 0 275 store volatile i64 %x, i64* %entry 276 ret void 277 } 278
