From 71002d8a5111d02ce8049c55017a8d948c820e35 Mon Sep 17 00:00:00 2001 From: Andreas Arnez Date: Thu, 25 Oct 2018 13:47:12 +0200 Subject: [PATCH] Bug 400490 s390x: Fix register allocation for VRs vs FPRs On s390x, if vector registers are available, they are fed to the register allocator as if they were separate from the floating-point registers. But in fact the FPRs are embedded in the VRs. So for instance, if both f3 and v3 are allocated and used at the same time, corruption will result. This is fixed by offering only the non-overlapping VRs, v16 to v31, to the register allocator instead. --- NEWS | 1 + VEX/priv/host_s390_defs.c | 17 +++++++---------- 2 files changed, 8 insertions(+), 10 deletions(-) diff --git a/NEWS b/NEWS index 987d92857e..63287bb3b2 100644 --- a/NEWS +++ b/NEWS @@ -47,6 +47,7 @@ where XXXXXX is the bug number as listed below. 399301 Use inlined frames in Massif XTree output. 399322 Improve callgrind_annotate output +400490 s390x: VRs allocated as if separate from FPRs Release 3.14.0 (9 October 2018) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ diff --git a/VEX/priv/host_s390_defs.c b/VEX/priv/host_s390_defs.c index 6c22ac8430..98ac9384db 100644 --- a/VEX/priv/host_s390_defs.c +++ b/VEX/priv/host_s390_defs.c @@ -59,7 +59,6 @@ static UInt s390_tchain_load64_len(void); /* A mapping from register number to register index */ static Int gpr_index[16]; // GPR regno -> register index -static Int fpr_index[16]; // FPR regno -> register index static Int vr_index[32]; // VR regno -> register index HReg @@ -73,7 +72,7 @@ s390_hreg_gpr(UInt regno) HReg s390_hreg_fpr(UInt regno) { - Int ix = fpr_index[regno]; + Int ix = vr_index[regno]; vassert(ix >= 0); return mkHReg(/*virtual*/False, HRcFlt64, regno, ix); } @@ -463,11 +462,9 @@ getRRegUniverse_S390(void) RRegUniverse__init(ru); - /* Assign invalid values to the gpr/fpr/vr_index */ + /* Assign invalid values to the gpr/vr_index */ for (UInt i = 0; i < sizeof gpr_index / sizeof gpr_index[0]; ++i) gpr_index[i] = -1; - for (UInt i = 0; i < sizeof fpr_index / sizeof fpr_index[0]; ++i) - fpr_index[i] = -1; for (UInt i = 0; i < sizeof vr_index / sizeof vr_index[0]; ++i) vr_index[i] = -1; @@ -494,17 +491,17 @@ getRRegUniverse_S390(void) ru->allocable_start[HRcFlt64] = ru->size; for (UInt regno = 8; regno <= 15; ++regno) { - fpr_index[regno] = ru->size; + vr_index[regno] = ru->size; ru->regs[ru->size++] = s390_hreg_fpr(regno); } for (UInt regno = 0; regno <= 7; ++regno) { - fpr_index[regno] = ru->size; + vr_index[regno] = ru->size; ru->regs[ru->size++] = s390_hreg_fpr(regno); } ru->allocable_end[HRcFlt64] = ru->size - 1; ru->allocable_start[HRcVec128] = ru->size; - for (UInt regno = 0; regno <= 31; ++regno) { + for (UInt regno = 16; regno <= 31; ++regno) { vr_index[regno] = ru->size; ru->regs[ru->size++] = s390_hreg_vr(regno); } @@ -527,12 +524,12 @@ getRRegUniverse_S390(void) /* Sanity checking */ for (UInt i = 0; i < sizeof gpr_index / sizeof gpr_index[0]; ++i) vassert(gpr_index[i] >= 0); - for (UInt i = 0; i < sizeof fpr_index / sizeof fpr_index[0]; ++i) - vassert(fpr_index[i] >= 0); for (UInt i = 0; i < sizeof vr_index / sizeof vr_index[0]; ++i) vassert(vr_index[i] >= 0); initialised = True; + + RRegUniverse__check_is_sane(ru); return ru; } -- 2.47.2