mednafen: fix build, add ppc patches, altivec option

Previously, mednafen wouldn't even build because of the aclocal
related issues and also that the libcdio option hasn't existed
in it in ages. So fixed that + added an altivec option for ppc
(enabled by default on ppc64, otherwise disabled) and ppc64
elfv2 libco impl from bsnes/higan.
This commit is contained in:
q66 2019-07-24 01:26:15 +02:00 committed by Jürgen Buchmüller
parent 99dc52799b
commit 2e36f254e9
3 changed files with 371 additions and 8 deletions

View File

@ -0,0 +1,25 @@
This fixes arch detection on all ppc as well as some incorrect handling
of endianness on little endian ppc64.
--- configure.ac 2019-01-28 06:52:37.000000000 +0000
+++ configure.ac 2019-03-24 22:30:31.180000000 +0000
@@ -765,7 +765,7 @@
AM_CONDITIONAL(ARCH_X86_32, true)
;;
- powerpc)
+ powerpc*|ppc*)
AC_DEFINE([ARCH_POWERPC], [1], [Define if we are compiling for PPC architectures.])
AM_CONDITIONAL(ARCH_POWERPC, true)
--- src/snes/src/lib/nall/detect.hpp
+++ src/snes/src/lib/nall/detect.hpp
@@ -21,7 +21,7 @@
/* Endian detection */
-#if defined(__i386__) || defined(__amd64__) || defined(_M_IX86) || defined(_M_AMD64)
+#if defined(__i386__) || defined(__amd64__) || defined(_M_IX86) || defined(_M_AMD64) || defined(__LITTLE_ENDIAN__)
#define ARCH_LSB
#elif defined(__powerpc__) || defined(_M_PPC) || defined(__BIG_ENDIAN__)
#define ARCH_MSB

View File

@ -0,0 +1,325 @@
Note: this is adapted from an equivalent change in higan/bsnes.
Slightly modified for mednafen (no settings.h, different paths)
From ddf550c1438d60b893a4fc1da333e021ac0e3658 Mon Sep 17 00:00:00 2001
From: Shawn Anastasio <shawn@anastas.io>
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
--- src/snes/src/lib/libco/libco.c
+++ src/snes/src/lib/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
+++ src/snes/src/lib/libco/ppc64v2.c
@@ -0,0 +1,283 @@
+/**
+ * libco implementation for ppc64 elfv2.
+ *
+ * Written by Shawn Anastasio.
+ * Licensed under the ISC license.
+ */
+
+#define LIBCO_C
+#include "libco.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <assert.h>
+
+#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

View File

@ -1,24 +1,37 @@
# Template file for 'mednafen'
pkgname=mednafen
version=1.22.2
revision=1
revision=2
wrksrc="$pkgname"
build_style=gnu-configure
configure_args="--with-libcdio --with-libsndfile --enable-alsa"
hostmakedepends="pkg-config"
makedepends="libcdio-devel libsndfile-devel glu-devel SDL2-devel zlib-devel"
configure_args="--with-libsndfile --enable-alsa $(vopt_enable altivec)
ac_cv_sizeof_off_t=8"
hostmakedepends="automake pkg-config libtool gettext-devel"
makedepends="libsndfile-devel glu-devel SDL2-devel zlib-devel"
short_desc="Portable, argument(command-line)-driven multi-system emulator"
maintainer="Juan RP <xtraeme@voidlinux.org>"
license="GPL-2.0-or-later"
homepage="https://mednafen.github.io/"
distfiles="https://mednafen.github.io/releases/files/mednafen-${version}.tar.xz"
checksum=fad433ac694696d69ea38f6f4be1d0a6c1aa3609ec7f46ce75412be2f2df2f95
CFLAGS="-D_FILE_OFFSET_BITS=64"
CXXFLAGS="$CFLAGS"
configure_args+=" ac_cv_sizeof_off_t=8"
nopie=yes
build_options="altivec"
desc_option_altivec="Enable AltiVec support on PowerPC"
case "$XBPS_TARGET_MACHINE" in
ppc64*) build_options_default="altivec";;
*) ;;
esac
pre_configure() {
# otherwise it tries to call aclocal-1.14 which we don't have
./autogen.sh
export CFLAGS="$CFLAGS -D_FILE_OFFSET_BITS=64"
export CXXFLAGS="$CXXFLAGS -D_FILE_OFFSET_BITS=64"
}
pre_build() {
# XXX not sure what's going on with those asserts.
sed -i '/assert/d' src/hash/sha{1,256}.cpp