From 4a6d19b057140e4e2ea478aa665323741296b94a Mon Sep 17 00:00:00 2001 From: q66 Date: Wed, 24 Jul 2019 00:16:30 +0200 Subject: [PATCH] higan: add a new implementation of libco for ppc64 ELFv2 Void uses ELFv2, and thanks to this we no longer need to fall back to the slow sjlj portable implementation. Upstream: https://github.com/byuu/higan/pull/9 [ci skip] --- srcpkgs/higan/patches/ppc64-libco-elfv2.patch | 323 ++++++++++++++++++ srcpkgs/higan/patches/ppc64-no-asm.patch | 26 -- srcpkgs/higan/template | 2 +- 3 files changed, 324 insertions(+), 27 deletions(-) create mode 100644 srcpkgs/higan/patches/ppc64-libco-elfv2.patch delete mode 100644 srcpkgs/higan/patches/ppc64-no-asm.patch diff --git a/srcpkgs/higan/patches/ppc64-libco-elfv2.patch b/srcpkgs/higan/patches/ppc64-libco-elfv2.patch new file mode 100644 index 00000000000000..087193a12c7ae2 --- /dev/null +++ b/srcpkgs/higan/patches/ppc64-libco-elfv2.patch @@ -0,0 +1,323 @@ +From ddf550c1438d60b893a4fc1da333e021ac0e3658 Mon Sep 17 00:00:00 2001 +From: Shawn Anastasio +Date: Tue, 23 Jul 2019 15:59:03 -0500 +Subject: [PATCH] Implement ppc64 ELFv2 support in libco + +The existing ppc implementation in libco only supports +the ELFv1 ABI on PowerPC 64 and therefore can't be used on +Little Endian systems and Big Endian systems running ELFv2 +distros. + +This commit introduces a new implementation of the libco +API for ppc64 elfv2. It has been tested with bsnes and higan. +The original ppc implementation is maintained for non-ELFv2 +targets. +--- + libco/libco.c | 4 +- + libco/ppc64v2.c | 284 ++++++++++++++++++++++++++++++++++++++++++++++++ + 2 files changed, 287 insertions(+), 1 deletion(-) + create mode 100644 libco/ppc64v2.c + +diff --git a/libco/libco.c b/libco/libco.c +index de11fbe9..f5ee5d0a 100755 +--- libco/libco.c ++++ libco/libco.c +@@ -9,6 +9,8 @@ + #include "amd64.c" + #elif defined(__arm__) + #include "arm.c" ++ #elif defined(__powerpc64__) && defined(_CALL_ELF) && (_CALL_ELF == 2) ++ #include "ppc64v2.c" + #elif defined(_ARCH_PPC) + #include "ppc.c" + #elif defined(_WIN32) +diff --git a/libco/ppc64v2.c b/libco/ppc64v2.c +new file mode 100644 +index 00000000..8f733de2 +--- /dev/null ++++ libco/ppc64v2.c +@@ -0,0 +1,284 @@ ++/** ++ * libco implementation for ppc64 elfv2. ++ * ++ * Written by Shawn Anastasio. ++ * Licensed under the ISC license. ++ */ ++ ++#define LIBCO_C ++#include "libco.h" ++#include "settings.h" ++ ++#include ++#include ++#include ++ ++#ifdef __cplusplus ++extern "C" { ++#endif ++ ++struct ppc64_context { ++ // GPRs ++ uint64_t gprs[32]; ++ uint64_t lr; ++ uint64_t ccr; ++ ++ // FPRs ++ uint64_t fprs[32]; ++ ++#ifdef __ALTIVEC__ ++ // Altivec (VMX) ++ uint64_t vmx[24 /* 12 non-volatile * 2 */]; ++ uint32_t vrsave; ++#endif ++}; ++ ++static thread_local struct ppc64_context *context_running = 0; ++ ++#define MAX(x, y) ((x) > (y) ? (x) : (y)) ++#define ALIGN(ptr, x) ( (void *)( (uintptr_t)(ptr) & ~((x)-1) ) ) ++ ++#define MIN_STACK 0x10000lu ++#define MIN_STACK_FRAME 0x20lu ++#define STACK_ALIGN 0x10lu ++ ++void swap_context(struct ppc64_context *read, struct ppc64_context *write); ++__asm__( ++ ".text\n" ++ ".align 4\n" ++ ".type swap_context @function\n" ++ "swap_context:\n" ++ ".cfi_startproc\n" ++ ++ // Dump non-volatile and special GPRs ++ "std 1, 8(4)\n" ++ "std 2, 16(4)\n" ++ "std 12, 96(4)\n" ++ "std 13, 104(4)\n" ++ "std 14, 112(4)\n" ++ "std 15, 120(4)\n" ++ "std 16, 128(4)\n" ++ "std 17, 136(4)\n" ++ "std 18, 144(4)\n" ++ "std 19, 152(4)\n" ++ "std 20, 160(4)\n" ++ "std 21, 168(4)\n" ++ "std 22, 176(4)\n" ++ "std 23, 184(4)\n" ++ "std 24, 192(4)\n" ++ "std 25, 200(4)\n" ++ "std 26, 208(4)\n" ++ "std 27, 216(4)\n" ++ "std 28, 224(4)\n" ++ "std 29, 232(4)\n" ++ "std 30, 240(4)\n" ++ "std 31, 248(4)\n" ++ ++ // LR ++ "mflr 5\n" ++ "std 5, 256(4)\n" ++ ++ // CCR ++ "mfcr 5\n" ++ "std 5, 264(4)\n" ++ ++ // Dump non-volatile FPRs ++ "stfd 14, 384(4)\n" ++ "stfd 15, 392(4)\n" ++ "stfd 16, 400(4)\n" ++ "stfd 17, 408(4)\n" ++ "stfd 18, 416(4)\n" ++ "stfd 19, 424(4)\n" ++ "stfd 20, 432(4)\n" ++ "stfd 21, 440(4)\n" ++ "stfd 22, 448(4)\n" ++ "stfd 23, 456(4)\n" ++ "stfd 24, 464(4)\n" ++ "stfd 25, 472(4)\n" ++ "stfd 26, 480(4)\n" ++ "stfd 27, 488(4)\n" ++ "stfd 28, 496(4)\n" ++ "stfd 29, 504(4)\n" ++ "stfd 30, 512(4)\n" ++ "stfd 31, 520(4)\n" ++ ++#ifdef __ALTIVEC__ ++ // Dump non-volatile VMX registers ++ "li 5, 528\n" ++ "stvxl 20, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 21, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 22, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 23, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 24, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 25, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 26, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 27, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 28, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 29, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 30, 4, 5\n" ++ "addi 5, 5, 16\n" ++ "stvxl 31, 4, 5\n" ++ "addi 5, 5, 16\n" ++ ++ // VRSAVE ++ "mfvrsave 5\n" ++ "stw 5, 736(4)\n" ++#endif ++ ++ // Restore GPRs ++ "ld 1, 8(3)\n" ++ "ld 2, 16(3)\n" ++ "ld 12, 96(3)\n" ++ "ld 13, 104(3)\n" ++ "ld 14, 112(3)\n" ++ "ld 15, 120(3)\n" ++ "ld 16, 128(3)\n" ++ "ld 17, 136(3)\n" ++ "ld 18, 144(3)\n" ++ "ld 19, 152(3)\n" ++ "ld 20, 160(3)\n" ++ "ld 21, 168(3)\n" ++ "ld 22, 176(3)\n" ++ "ld 23, 184(3)\n" ++ "ld 24, 192(3)\n" ++ "ld 25, 200(3)\n" ++ "ld 26, 208(3)\n" ++ "ld 27, 216(3)\n" ++ "ld 28, 224(3)\n" ++ "ld 29, 232(3)\n" ++ "ld 30, 240(3)\n" ++ "ld 31, 248(3)\n" ++ ++ // Restore LR ++ "ld 5, 256(3)\n" ++ "mtlr 5\n" ++ ++ // Restore CCR ++ "ld 5, 264(3)\n" ++ "mtcr 5\n" ++ ++ // Restore FPRs ++ "lfd 14, 384(3)\n" ++ "lfd 15, 392(3)\n" ++ "lfd 16, 400(3)\n" ++ "lfd 17, 408(3)\n" ++ "lfd 18, 416(3)\n" ++ "lfd 19, 424(3)\n" ++ "lfd 20, 432(3)\n" ++ "lfd 21, 440(3)\n" ++ "lfd 22, 448(3)\n" ++ "lfd 23, 456(3)\n" ++ "lfd 24, 464(3)\n" ++ "lfd 25, 472(3)\n" ++ "lfd 26, 480(3)\n" ++ "lfd 27, 488(3)\n" ++ "lfd 28, 496(3)\n" ++ "lfd 29, 504(3)\n" ++ "lfd 30, 512(3)\n" ++ "lfd 31, 520(3)\n" ++ ++#ifdef __ALTIVEC__ ++ // Restore VMX ++ "li 5, 528\n" ++ "lvxl 20, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 21, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 22, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 23, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 24, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 25, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 26, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 27, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 28, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 29, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 30, 3, 5\n" ++ "addi 5, 5, 16\n" ++ "lvxl 31, 3, 5\n" ++ "addi 5, 5, 16\n" ++ ++ // VRSAVE ++ "lwz 5, 720(3)\n" ++ "mtvrsave 5\n" ++#endif ++ ++ // Context restored, branch to LR ++ "blr\n" ++ ++ ".cfi_endproc\n" ++ ".size swap_context, .-swap_context\n" ++); ++ ++cothread_t co_active() { ++ if (!context_running) ++ context_running = (struct ppc64_context *) ++ malloc(MIN_STACK + sizeof(struct ppc64_context)); ++ return (cothread_t)context_running; ++} ++ ++cothread_t co_derive(void *memory, unsigned int size, void (*coentry)(void)) { ++ uint8_t *sp; ++ struct ppc64_context *context = (struct ppc64_context *)memory; ++ ++ // Save current context into new context to initialize it ++ swap_context(context, context); ++ ++ // Align stack ++ sp = (uint8_t *)memory + size - STACK_ALIGN; ++ sp = (uint8_t *)ALIGN(sp, STACK_ALIGN); ++ ++ // Write 0 for initial backchain ++ *(uint64_t *)sp = 0; ++ ++ // Create new frame with backchain ++ sp -= MIN_STACK_FRAME; ++ *(uint64_t *)sp = (uint64_t)(sp + MIN_STACK_FRAME); ++ ++ // Update context with new stack (r1) and entrypoint (LR, r12) ++ context->lr = (uint64_t)coentry; ++ context->gprs[12] = (uint64_t)coentry; ++ context->gprs[1] = (uint64_t)sp; ++ ++ return (cothread_t)memory; ++} ++ ++cothread_t co_create(unsigned int size, void (*coentry)(void)) { ++ size_t total = MAX(size, MIN_STACK) + sizeof(struct ppc64_context); ++ void *memory = malloc(total); ++ if (!memory) ++ return (cothread_t)0; ++ ++ return co_derive(memory, total, coentry); ++} ++ ++void co_delete(cothread_t t) { ++ free(t); ++} ++ ++void co_switch(cothread_t t) { ++ struct ppc64_context *old = context_running; ++ context_running = (struct ppc64_context *)t; ++ swap_context((struct ppc64_context *)t, old); ++} ++ ++#ifdef __cplusplus ++} ++#endif diff --git a/srcpkgs/higan/patches/ppc64-no-asm.patch b/srcpkgs/higan/patches/ppc64-no-asm.patch deleted file mode 100644 index e2a58d5f3df6b2..00000000000000 --- a/srcpkgs/higan/patches/ppc64-no-asm.patch +++ /dev/null @@ -1,26 +0,0 @@ -Higan's ppc code assumes big endian, so fall back to sjlj for that. ELFv2 big -endian just needs a small patch to function. - ---- libco/libco.c -+++ libco/libco.c -@@ -9,7 +9,7 @@ - #include "amd64.c" - #elif defined(__arm__) - #include "arm.c" -- #elif defined(_ARCH_PPC) -+ #elif defined(_ARCH_PPC) && !defined(__LITTLE_ENDIAN__) - #include "ppc.c" - #elif defined(_WIN32) - #include "fiber.c" ---- libco/ppc.c -+++ libco/ppc.c -@@ -36,7 +36,8 @@ static thread_local cothread_t co_active_handle = 0; - - /* whether function calls are indirect through a descriptor, or are directly to function */ - #ifndef LIBCO_PPCDESC -- #if !_CALL_SYSV && (_CALL_AIX || _CALL_AIXDESC || LIBCO_PPC64) -+ #if !_CALL_SYSV && (_CALL_AIX || _CALL_AIXDESC || \ -+ (LIBCO_PPC64 && (!defined(_CALL_ELF) || _CALL_ELF == 1))) - #define LIBCO_PPCDESC 1 - #endif - #endif diff --git a/srcpkgs/higan/template b/srcpkgs/higan/template index 94bbb2afebbe8c..0835499f205f2e 100644 --- a/srcpkgs/higan/template +++ b/srcpkgs/higan/template @@ -3,8 +3,8 @@ pkgname=higan version=106 revision=1 wrksrc="higan_v${version}-source" -build_style=gnu-makefile build_wrksrc=higan +build_style=gnu-makefile make_build_args='compiler=${CXX}' make_install_args="prefix=${DESTDIR}/usr" hostmakedepends="p7zip pkg-config SDL-devel"