1; Test the saving and restoring of FPRs in large frames. 2; 3; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck -check-prefix=CHECK-NOFP %s 4; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 -disable-fp-elim | FileCheck -check-prefix=CHECK-FP %s 5 6; Test a frame size that requires some FPRs to be saved and loaded using 7; the 20-bit STDY and LDY while others can use the 12-bit STD and LD. 8; The frame is big enough to require two emergency spill slots at 160(%r15), 9; as well as the 8 FPR save slots. Get a frame of size 4128 by allocating 10; (4128 - 176 - 8 * 8) / 8 = 486 extra doublewords. 11define void @f1(double *%ptr, i64 %x) { 12; CHECK-NOFP-LABEL: f1: 13; CHECK-NOFP: aghi %r15, -4128 14; CHECK-NOFP: .cfi_def_cfa_offset 4288 15; CHECK-NOFP: stdy %f8, 4120(%r15) 16; CHECK-NOFP: stdy %f9, 4112(%r15) 17; CHECK-NOFP: stdy %f10, 4104(%r15) 18; CHECK-NOFP: stdy %f11, 4096(%r15) 19; CHECK-NOFP: std %f12, 4088(%r15) 20; CHECK-NOFP: std %f13, 4080(%r15) 21; CHECK-NOFP: std %f14, 4072(%r15) 22; CHECK-NOFP: std %f15, 4064(%r15) 23; CHECK-NOFP: .cfi_offset %f8, -168 24; CHECK-NOFP: .cfi_offset %f9, -176 25; CHECK-NOFP: .cfi_offset %f10, -184 26; CHECK-NOFP: .cfi_offset %f11, -192 27; CHECK-NOFP: .cfi_offset %f12, -200 28; CHECK-NOFP: .cfi_offset %f13, -208 29; CHECK-NOFP: .cfi_offset %f14, -216 30; CHECK-NOFP: .cfi_offset %f15, -224 31; ...main function body... 32; CHECK-NOFP: ldy %f8, 4120(%r15) 33; CHECK-NOFP: ldy %f9, 4112(%r15) 34; CHECK-NOFP: ldy %f10, 4104(%r15) 35; CHECK-NOFP: ldy %f11, 4096(%r15) 36; CHECK-NOFP: ld %f12, 4088(%r15) 37; CHECK-NOFP: ld %f13, 4080(%r15) 38; CHECK-NOFP: ld %f14, 4072(%r15) 39; CHECK-NOFP: ld %f15, 4064(%r15) 40; CHECK-NOFP: aghi %r15, 4128 41; CHECK-NOFP: br %r14 42; 43; CHECK-FP-LABEL: f1: 44; CHECK-FP: stmg %r11, %r15, 88(%r15) 45; CHECK-FP: aghi %r15, -4128 46; CHECK-FP: .cfi_def_cfa_offset 4288 47; CHECK-FP: lgr %r11, %r15 48; CHECK-FP: .cfi_def_cfa_register %r11 49; CHECK-FP: stdy %f8, 4120(%r11) 50; CHECK-FP: stdy %f9, 4112(%r11) 51; CHECK-FP: stdy %f10, 4104(%r11) 52; CHECK-FP: stdy %f11, 4096(%r11) 53; CHECK-FP: std %f12, 4088(%r11) 54; CHECK-FP: std %f13, 4080(%r11) 55; CHECK-FP: std %f14, 4072(%r11) 56; CHECK-FP: std %f15, 4064(%r11) 57; ...main function body... 58; CHECK-FP: ldy %f8, 4120(%r11) 59; CHECK-FP: ldy %f9, 4112(%r11) 60; CHECK-FP: ldy %f10, 4104(%r11) 61; CHECK-FP: ldy %f11, 4096(%r11) 62; CHECK-FP: ld %f12, 4088(%r11) 63; CHECK-FP: ld %f13, 4080(%r11) 64; CHECK-FP: ld %f14, 4072(%r11) 65; CHECK-FP: ld %f15, 4064(%r11) 66; CHECK-FP: lmg %r11, %r15, 4216(%r11) 67; CHECK-FP: br %r14 68 %y = alloca [486 x i64], align 8 69 %elem = getelementptr inbounds [486 x i64], [486 x i64]* %y, i64 0, i64 0 70 store volatile i64 %x, i64* %elem 71 %l0 = load volatile double , double *%ptr 72 %l1 = load volatile double , double *%ptr 73 %l2 = load volatile double , double *%ptr 74 %l3 = load volatile double , double *%ptr 75 %l4 = load volatile double , double *%ptr 76 %l5 = load volatile double , double *%ptr 77 %l6 = load volatile double , double *%ptr 78 %l7 = load volatile double , double *%ptr 79 %l8 = load volatile double , double *%ptr 80 %l9 = load volatile double , double *%ptr 81 %l10 = load volatile double , double *%ptr 82 %l11 = load volatile double , double *%ptr 83 %l12 = load volatile double , double *%ptr 84 %l13 = load volatile double , double *%ptr 85 %l14 = load volatile double , double *%ptr 86 %l15 = load volatile double , double *%ptr 87 %add0 = fadd double %l0, %l0 88 %add1 = fadd double %l1, %add0 89 %add2 = fadd double %l2, %add1 90 %add3 = fadd double %l3, %add2 91 %add4 = fadd double %l4, %add3 92 %add5 = fadd double %l5, %add4 93 %add6 = fadd double %l6, %add5 94 %add7 = fadd double %l7, %add6 95 %add8 = fadd double %l8, %add7 96 %add9 = fadd double %l9, %add8 97 %add10 = fadd double %l10, %add9 98 %add11 = fadd double %l11, %add10 99 %add12 = fadd double %l12, %add11 100 %add13 = fadd double %l13, %add12 101 %add14 = fadd double %l14, %add13 102 %add15 = fadd double %l15, %add14 103 store volatile double %add0, double *%ptr 104 store volatile double %add1, double *%ptr 105 store volatile double %add2, double *%ptr 106 store volatile double %add3, double *%ptr 107 store volatile double %add4, double *%ptr 108 store volatile double %add5, double *%ptr 109 store volatile double %add6, double *%ptr 110 store volatile double %add7, double *%ptr 111 store volatile double %add8, double *%ptr 112 store volatile double %add9, double *%ptr 113 store volatile double %add10, double *%ptr 114 store volatile double %add11, double *%ptr 115 store volatile double %add12, double *%ptr 116 store volatile double %add13, double *%ptr 117 store volatile double %add14, double *%ptr 118 store volatile double %add15, double *%ptr 119 ret void 120} 121 122; Test a frame size that requires some FPRs to be saved and loaded using 123; an indexed STD and LD while others can use the 20-bit STDY and LDY. 124; The index can be any call-clobbered GPR except %r0. 125; 126; Don't require the accesses to share the same LLILH; that would be a 127; good optimisation but is really a different test. 128; 129; As above, get a frame of size 524320 by allocating 130; (524320 - 176 - 8 * 8) / 8 = 65510 extra doublewords. 131define void @f2(double *%ptr, i64 %x) { 132; CHECK-NOFP-LABEL: f2: 133; CHECK-NOFP: agfi %r15, -524320 134; CHECK-NOFP: .cfi_def_cfa_offset 524480 135; CHECK-NOFP: llilh [[INDEX:%r[1-5]]], 8 136; CHECK-NOFP: std %f8, 24([[INDEX]],%r15) 137; CHECK-NOFP: std %f9, 16({{%r[1-5]}},%r15) 138; CHECK-NOFP: std %f10, 8({{%r[1-5]}},%r15) 139; CHECK-NOFP: std %f11, 0({{%r[1-5]}},%r15) 140; CHECK-NOFP: stdy %f12, 524280(%r15) 141; CHECK-NOFP: stdy %f13, 524272(%r15) 142; CHECK-NOFP: stdy %f14, 524264(%r15) 143; CHECK-NOFP: stdy %f15, 524256(%r15) 144; CHECK-NOFP: .cfi_offset %f8, -168 145; CHECK-NOFP: .cfi_offset %f9, -176 146; CHECK-NOFP: .cfi_offset %f10, -184 147; CHECK-NOFP: .cfi_offset %f11, -192 148; CHECK-NOFP: .cfi_offset %f12, -200 149; CHECK-NOFP: .cfi_offset %f13, -208 150; CHECK-NOFP: .cfi_offset %f14, -216 151; CHECK-NOFP: .cfi_offset %f15, -224 152; ...main function body... 153; CHECK-NOFP: ld %f8, 24({{%r[1-5]}},%r15) 154; CHECK-NOFP: ld %f9, 16({{%r[1-5]}},%r15) 155; CHECK-NOFP: ld %f10, 8({{%r[1-5]}},%r15) 156; CHECK-NOFP: ld %f11, 0({{%r[1-5]}},%r15) 157; CHECK-NOFP: ldy %f12, 524280(%r15) 158; CHECK-NOFP: ldy %f13, 524272(%r15) 159; CHECK-NOFP: ldy %f14, 524264(%r15) 160; CHECK-NOFP: ldy %f15, 524256(%r15) 161; CHECK-NOFP: agfi %r15, 524320 162; CHECK-NOFP: br %r14 163; 164; CHECK-FP-LABEL: f2: 165; CHECK-FP: stmg %r11, %r15, 88(%r15) 166; CHECK-FP: agfi %r15, -524320 167; CHECK-FP: .cfi_def_cfa_offset 524480 168; CHECK-FP: llilh [[INDEX:%r[1-5]]], 8 169; CHECK-FP: std %f8, 24([[INDEX]],%r11) 170; CHECK-FP: std %f9, 16({{%r[1-5]}},%r11) 171; CHECK-FP: std %f10, 8({{%r[1-5]}},%r11) 172; CHECK-FP: std %f11, 0({{%r[1-5]}},%r11) 173; CHECK-FP: stdy %f12, 524280(%r11) 174; CHECK-FP: stdy %f13, 524272(%r11) 175; CHECK-FP: stdy %f14, 524264(%r11) 176; CHECK-FP: stdy %f15, 524256(%r11) 177; CHECK-FP: .cfi_offset %f8, -168 178; CHECK-FP: .cfi_offset %f9, -176 179; CHECK-FP: .cfi_offset %f10, -184 180; CHECK-FP: .cfi_offset %f11, -192 181; CHECK-FP: .cfi_offset %f12, -200 182; CHECK-FP: .cfi_offset %f13, -208 183; CHECK-FP: .cfi_offset %f14, -216 184; CHECK-FP: .cfi_offset %f15, -224 185; ...main function body... 186; CHECK-FP: ld %f8, 24({{%r[1-5]}},%r11) 187; CHECK-FP: ld %f9, 16({{%r[1-5]}},%r11) 188; CHECK-FP: ld %f10, 8({{%r[1-5]}},%r11) 189; CHECK-FP: ld %f11, 0({{%r[1-5]}},%r11) 190; CHECK-FP: ldy %f12, 524280(%r11) 191; CHECK-FP: ldy %f13, 524272(%r11) 192; CHECK-FP: ldy %f14, 524264(%r11) 193; CHECK-FP: ldy %f15, 524256(%r11) 194; CHECK-FP: aghi %r11, 128 195; CHECK-FP: lmg %r11, %r15, 524280(%r11) 196; CHECK-FP: br %r14 197 %y = alloca [65510 x i64], align 8 198 %elem = getelementptr inbounds [65510 x i64], [65510 x i64]* %y, i64 0, i64 0 199 store volatile i64 %x, i64* %elem 200 %l0 = load volatile double , double *%ptr 201 %l1 = load volatile double , double *%ptr 202 %l2 = load volatile double , double *%ptr 203 %l3 = load volatile double , double *%ptr 204 %l4 = load volatile double , double *%ptr 205 %l5 = load volatile double , double *%ptr 206 %l6 = load volatile double , double *%ptr 207 %l7 = load volatile double , double *%ptr 208 %l8 = load volatile double , double *%ptr 209 %l9 = load volatile double , double *%ptr 210 %l10 = load volatile double , double *%ptr 211 %l11 = load volatile double , double *%ptr 212 %l12 = load volatile double , double *%ptr 213 %l13 = load volatile double , double *%ptr 214 %l14 = load volatile double , double *%ptr 215 %l15 = load volatile double , double *%ptr 216 %add0 = fadd double %l0, %l0 217 %add1 = fadd double %l1, %add0 218 %add2 = fadd double %l2, %add1 219 %add3 = fadd double %l3, %add2 220 %add4 = fadd double %l4, %add3 221 %add5 = fadd double %l5, %add4 222 %add6 = fadd double %l6, %add5 223 %add7 = fadd double %l7, %add6 224 %add8 = fadd double %l8, %add7 225 %add9 = fadd double %l9, %add8 226 %add10 = fadd double %l10, %add9 227 %add11 = fadd double %l11, %add10 228 %add12 = fadd double %l12, %add11 229 %add13 = fadd double %l13, %add12 230 %add14 = fadd double %l14, %add13 231 %add15 = fadd double %l15, %add14 232 store volatile double %add0, double *%ptr 233 store volatile double %add1, double *%ptr 234 store volatile double %add2, double *%ptr 235 store volatile double %add3, double *%ptr 236 store volatile double %add4, double *%ptr 237 store volatile double %add5, double *%ptr 238 store volatile double %add6, double *%ptr 239 store volatile double %add7, double *%ptr 240 store volatile double %add8, double *%ptr 241 store volatile double %add9, double *%ptr 242 store volatile double %add10, double *%ptr 243 store volatile double %add11, double *%ptr 244 store volatile double %add12, double *%ptr 245 store volatile double %add13, double *%ptr 246 store volatile double %add14, double *%ptr 247 store volatile double %add15, double *%ptr 248 ret void 249} 250