main/luajit: ppc64: fix r12 TOC issue and remove workaround

2 files changed, 47 insertions, 24 deletions

diff --git a/main/luajit/APKBUILD b/main/luajit/APKBUILD
index 77bd453dd2..b554116c4b 100644
--- a/main/luajit/APKBUILD
+++ b/main/luajit/APKBUILD
@@ -4,8 +4,8 @@
 pkgname=luajit
 pkgver=2.1.0_beta2
 _realver=${pkgver/_/-}
-pkgrel=3
-pkgdesc='Just-in-time compiler and replacement for Lua 5.1 '
+pkgrel=4
+pkgdesc='Just-in-time compiler and replacement for Lua 5.1'
 url='http://luajit.org'
 arch="all !s390x"
 license="MIT"
@@ -18,9 +18,6 @@ builddir=$srcdir/LuaJIT-$_realver
 
 build() {
 	cd "$builddir"
-	if [ "$CARCH" = "ppc64le" ]; then
-		local LDFLAGS="-no-pie"
-	fi
 	make amalg PREFIX=/usr || return 1
 }
 
@@ -39,4 +36,4 @@ package() {
 }
 
 sha512sums="ef04db5bee66a6e24da8d488df2c035c9593792dca4cb49955d673281a7c7b00ab40f4a63e699959461da7fb0732cf81ec46808c33b66ae1eaca4ef94d0b61e3  LuaJIT-2.1.0-beta2.tar.gz
-f9823c3a5091338961e1e70a47e30d0d3e592cdf8d085405d3201a1739cb1ebaaca02eec8ee9a88eab3d80466f5f4227a847553b15f9dbb209fe789ae3b37b67  enable-support-for-ppc64le.patch"
+d48d74c856aed035d17f8355178cab4c17b5fed97d0a218fee7293fbb5c262b564cbc4ecf2e7c8558186371ae57df6f932e0313ed0453552ef99052f82876241  enable-support-for-ppc64le.patch"
diff --git a/main/luajit/enable-support-for-ppc64le.patch b/main/luajit/enable-support-for-ppc64le.patch
index 06be3f8251..0cd000b201 100644
--- a/main/luajit/enable-support-for-ppc64le.patch
+++ b/main/luajit/enable-support-for-ppc64le.patch
@@ -1,4 +1,4 @@
-From e54d7b18e777a05872e2aaa4ab40041a2c6db768 Mon Sep 17 00:00:00 2001
+From d2400df1c753683571c5df9ea29afb37b6583d36 Mon Sep 17 00:00:00 2001
 From: Gustavo Serra Scalet <gsscalet@gmail.com>
 Date: Tue, 2 Jun 2015 14:36:20 -0300
 Subject: [PATCH] PPC64: Enable support for ppc64 little endian
@@ -7,13 +7,14 @@ Subject: [PATCH] PPC64: Enable support for ppc64 little endian
  dynasm/dasm_ppc.lua    |   11 +-
  src/host/buildvm_asm.c |   19 +-
  src/lj_arch.h          |    4 +-
- src/lj_ccall.c         |   32 +
+ src/lj_ccall.c         |   37 +
  src/lj_ccall.h         |    9 +
  src/lj_ccallback.c     |   15 +
+ src/lj_ctype.h         |    2 +-
  src/lj_frame.h         |    9 +
  src/lj_target_ppc.h    |   11 +-
- src/vm_ppc64.dasc      | 4508 ++++++++++++++++++++++++++++++++++++++++++++++++
- 9 files changed, 4608 insertions(+), 10 deletions(-)
+ src/vm_ppc64.dasc      | 4515 ++++++++++++++++++++++++++++++++++++++++++++++++
+ 10 files changed, 4621 insertions(+), 11 deletions(-)
  create mode 100644 src/vm_ppc64.dasc
 
 diff --git a/dynasm/dasm_ppc.lua b/dynasm/dasm_ppc.lua
@@ -110,10 +111,10 @@ index f1e7d7f..e2cdda7 100644
  #ifdef __NO_FPRS__
  #error "No support for PPC/e500 anymore (use LuaJIT 2.0)"
 diff --git a/src/lj_ccall.c b/src/lj_ccall.c
-index 5ab5b60..6b01544 100644
+index 5ab5b60..9ecad9d 100644
 --- a/src/lj_ccall.c
 +++ b/src/lj_ccall.c
-@@ -380,6 +380,37 @@
+@@ -380,6 +380,42 @@
  #define CCALL_HANDLE_COMPLEXARG \
    /* Pass complex by value in 2 or 4 GPRs. */
  
@@ -127,8 +128,13 @@ index 5ab5b60..6b01544 100644
 +      if (nfpr + 1 <= CCALL_NARG_FPR) { \
 +	dp = &cc->fpr[nfpr]; \
 +	nfpr += 1; \
-+	ngpr += 1;  /* align GPRs */ \
 +	d = ctype_get(cts, CTID_DOUBLE);  /* FPRs always hold doubles. */ \
++	if (ngpr + 1 <= maxgpr) \
++	  ngpr += 1;  /* align GPRs */ \
++	else if (nsp + 1 <= CCALL_MAXSTACK) \
++	  nsp += 1; /* align save area slots */ \
++	else \
++	  goto err_nyi; /* Too many args */ \
 +	goto done; \
 +      } \
 +    } else {  /* Try to pass argument in GPRs. */ \
@@ -151,7 +157,7 @@ index 5ab5b60..6b01544 100644
  #define CCALL_HANDLE_REGARG \
    if (isfp) {  /* Try to pass argument in FPRs. */ \
      if (nfpr + 1 <= CCALL_NARG_FPR) { \
-@@ -402,6 +433,7 @@
+@@ -402,6 +438,7 @@
        goto done; \
      } \
    }
@@ -160,7 +166,7 @@ index 5ab5b60..6b01544 100644
  #define CCALL_HANDLE_RET \
    if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \
 diff --git a/src/lj_ccall.h b/src/lj_ccall.h
-index 91983fe..8ab50d2 100644
+index 91983fe..0994d84 100644
 --- a/src/lj_ccall.h
 +++ b/src/lj_ccall.h
 @@ -85,12 +85,21 @@ typedef union FPRArg {
@@ -169,7 +175,7 @@ index 91983fe..8ab50d2 100644
  
 +#if LJ_ARCH_PPC64
 +#define CCALL_NARG_GPR		8
-+#define CCALL_NARG_FPR		8
++#define CCALL_NARG_FPR		13
 +#define CCALL_NRET_GPR		4	/* For complex double. */
 +#define CCALL_NRET_FPR		1
 +#define CCALL_SPS_EXTRA		14
@@ -223,6 +229,19 @@ index 065c329..0ef527b 100644
    *p++ = PPCI_MTCTR | PPCF_T(RID_TMP);
    *p++ = PPCI_BCTR;
    for (slot = 0; slot < CALLBACK_MAX_SLOT; slot++) {
+diff --git a/src/lj_ctype.h b/src/lj_ctype.h
+index 6639547..2f09d7d 100644
+--- a/src/lj_ctype.h
++++ b/src/lj_ctype.h
+@@ -153,7 +153,7 @@ typedef struct CType {
+ 
+ /* Simplify target-specific configuration. Checked in lj_ccall.h. */
+ #define CCALL_MAX_GPR		8
+-#define CCALL_MAX_FPR		8
++#define CCALL_MAX_FPR		13
+ 
+ typedef LJ_ALIGN(8) union FPRCBArg { double d; float f[2]; } FPRCBArg;
+ 
 diff --git a/src/lj_frame.h b/src/lj_frame.h
 index a86c36b..7ec6478 100644
 --- a/src/lj_frame.h
@@ -279,10 +298,10 @@ index 9986768..7a68e2a 100644
    PPCI_BC = 0x40800000,
 diff --git a/src/vm_ppc64.dasc b/src/vm_ppc64.dasc
 new file mode 100644
-index 0000000..45e5af9
+index 0000000..d7e6bb6
 --- /dev/null
 +++ b/src/vm_ppc64.dasc
-@@ -0,0 +1,4508 @@
+@@ -0,0 +1,4515 @@
 +|// Low-level VM code for PPC64 CPUs.
 +|// Bytecode interpreter, fast functions and helper functions.
 +|// Copyright (C) 2005-2015 Mike Pall. See Copyright Notice in luajit.h
@@ -344,7 +363,7 @@ index 0000000..45e5af9
 +|.define RA,		r20	// Callee-save.
 +|.define RB,		r10
 +|.define RC,		r11
-+|.define RD,		r12
++|.define RD,		r12	// Also used as function linkage register
 +|.define INS,		r7	// Overlaps CARG5.
 +|
 +|.define TMP0,		r0
@@ -395,7 +414,7 @@ index 0000000..45e5af9
 +|.define SAVE_TOC,	24(sp)	// TOC save area.
 +|// Next frame lr:	16(sp)
 +|.define SAVE_CR,	8(sp)	// 64 bit CR save.
-+|// Back chain for sp:	0(sp) 	<-- sp while in interpreter
++|// Back chain for sp:	0(sp)   <-- sp while in interpreter
 +|
 +|.define TMPD_BLO,	40(sp)	// LSB
 +|.define TMPD,		TMPD_LO	// base address of TMPD doubleword
@@ -981,7 +1000,8 @@ index 0000000..45e5af9
 +  |    std TMP1, SAVE_CFRAME
 +  |    std sp, L->cframe		// Add our C frame to cframe chain.
 +  |     std L, DISPATCH_GL(cur_L)(DISPATCH)
-+  |  mtctr CARG4
++  |  mr r12, CARG4			// keep r12 for function linkage.
++  |  mtctr r12
 +  |  bctrl			// (lua_State *L, lua_CFunction func, void *ud)
 +  |  mr. BASE, CRET1
 +  |   li PC, FRAME_CP
@@ -2344,7 +2364,8 @@ index 0000000..45e5af9
 +  |    std TMP1, L->top
 +  |   mr CARG1, L
 +  |  bgt >5				// Need to grow stack.
-+  |  mtctr TMP3
++  |  mr r12, TMP3			// keep r12 for function linkage.
++  |  mtctr r12
 +  |  bctrl				// (lua_State *L)
 +  |  // Either throws an error, or recovers and returns -1, 0 or nresults+1.
 +  |  ld BASE, L->base
@@ -2677,6 +2698,11 @@ index 0000000..45e5af9
 +  |  lfd f6, CCSTATE->fpr[5]
 +  |  lfd f7, CCSTATE->fpr[6]
 +  |  lfd f8, CCSTATE->fpr[7]
++  |  lfd f9, CCSTATE->fpr[8]
++  |  lfd f10, CCSTATE->fpr[9]
++  |  lfd f11, CCSTATE->fpr[10]
++  |  lfd f12, CCSTATE->fpr[11]
++  |  lfd f13, CCSTATE->fpr[12]
 +  |3:
 +  |  ld r12, CCSTATE->func
 +  |  ld CARG2, CCSTATE->gpr[1]
@@ -4611,7 +4637,7 @@ index 0000000..45e5af9
 +    |   cmpld TMP1, TMP2
 +    |    std RC, L->top
 +    |     li_vmstate C
-+    |  mtctr RD
++    |  mtctr RD				// RD is r12, the function linkage register
 +    if (op == BC_FUNCCW) {
 +      |  ld CARG2, CFUNC:RB->f
 +    }
@@ -4792,5 +4818,5 @@ index 0000000..45e5af9
 +  }
 +}
 -- 
-2.11.0
+2.12.2