new_dynarec: autodetect ARM integer divide feature at runtime

jni/Android.mk

@@ -134,6 +134,7 @@ LOCAL_LDLIBS := -lz
 
 ifeq ($(TARGET_ARCH_ABI), armeabi-v7a)
     # Use for ARM7a:
+    LOCAL_SRC_FILES += $(SRCDIR)/r4300/new_dynarec/arm_cpu_features.cpp
     LOCAL_SRC_FILES += $(SRCDIR)/r4300/new_dynarec/linkage_arm.S
     LOCAL_CFLAGS += -DDYNAREC
     LOCAL_CFLAGS += -DNEW_DYNAREC=3
@@ -142,6 +143,7 @@ ifeq ($(TARGET_ARCH_ABI), armeabi-v7a)
 
 else ifeq ($(TARGET_ARCH_ABI), armeabi)
     # Use for pre-ARM7a:
+    LOCAL_SRC_FILES += $(SRCDIR)/r4300/new_dynarec/arm_cpu_features.cpp
     LOCAL_SRC_FILES += $(SRCDIR)/r4300/new_dynarec/linkage_arm.S
     LOCAL_CFLAGS += -DARMv5_ONLY
     LOCAL_CFLAGS += -DDYNAREC

jni/mupen64plus-core/src/r4300/new_dynarec/arm_cpu_features.cpp

@@ -0,0 +1,150 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - arm_cpu_features.cpp                                    *
+ *   Mupen64Plus homepage: http://code.google.com/p/mupen64plus/           *
+ *   Copyright (C) 2015 Gilles Siberlin                                    *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+#include <fstream>
+#include <sstream>
+#include <string>
+#include "arm_cpu_features.h"
+
+extern "C" {
+    #include "api/callbacks.h"
+}
+
+arm_cpu_features_t arm_cpu_features;
+
+const char procfile[] = "/proc/cpuinfo";
+
+static unsigned char check_arm_cpu_feature(const std::string& feature)
+{
+    const std::string marker = "Features\t: ";
+
+    std::string line;
+    std::ifstream file(procfile);
+
+    if (!file)
+        return 0;
+
+    while (std::getline(file, line))
+    {
+        if (line.find(marker) != std::string::npos)
+        {
+            std::stringstream line_stream(line);
+            std::string token;
+
+            while (std::getline(line_stream, token, ' '))
+            {
+                if (token == feature)
+                    return 1;
+            }
+        }
+    }
+    return 0;
+}
+
+static unsigned char get_arm_cpu_implementer(void)
+{
+    const std::string marker = "CPU implementer\t: ";
+    unsigned char implementer = 0;
+
+    std::string line;
+    std::ifstream file(procfile);
+
+    if (!file)
+        return 0;
+
+    while (std::getline(file, line))
+    {
+        if (line.find(marker) != std::string::npos)
+        {
+            line = line.substr(marker.length());
+            sscanf(line.c_str(), "0x%02hhx", &implementer);
+            break;
+        }
+    }
+    return implementer;
+}
+
+static unsigned short get_arm_cpu_part(void)
+{
+    const std::string marker = "CPU part\t: ";
+    unsigned short part = 0;
+
+    std::string line;
+    std::ifstream file(procfile);
+
+    if (!file)
+        return 0;
+
+    while (std::getline(file, line))
+    {
+        if (line.find(marker) != std::string::npos)
+        {
+            line = line.substr(marker.length());
+            sscanf(line.c_str(), "0x%03hx", &part);
+            break;
+        }
+    }
+    return part;
+}
+
+void detect_arm_cpu_features(void)
+{
+    arm_cpu_features.SWP = check_arm_cpu_feature("swp");
+    arm_cpu_features.Half = check_arm_cpu_feature("half");
+    arm_cpu_features.Thumb = check_arm_cpu_feature("thumb");
+    arm_cpu_features.FastMult = check_arm_cpu_feature("fastmult");
+    arm_cpu_features.VFP = check_arm_cpu_feature("vfp");
+    arm_cpu_features.EDSP = check_arm_cpu_feature("edsp");
+    arm_cpu_features.ThumbEE = check_arm_cpu_feature("thumbee");
+    arm_cpu_features.NEON = check_arm_cpu_feature("neon");
+    arm_cpu_features.VFPv3 = check_arm_cpu_feature("vfpv3");
+    arm_cpu_features.TLS = check_arm_cpu_feature("tls");
+    arm_cpu_features.VFPv4 = check_arm_cpu_feature("vfpv4");
+    arm_cpu_features.IDIVa = check_arm_cpu_feature("idiva");
+    arm_cpu_features.IDIVt = check_arm_cpu_feature("idivt");
+    // Qualcomm Krait supports IDIVa but it doesn't report it. Check for krait.
+    if (get_arm_cpu_implementer() == 0x51 && get_arm_cpu_part() == 0x6F)
+        arm_cpu_features.IDIVa = arm_cpu_features.IDIVt = 1;
+}
+
+void print_arm_cpu_features(void)
+{
+    std::string arm_cpu_features_string;
+
+    arm_cpu_features_string="ARM CPU Features:";
+
+    if (arm_cpu_features.SWP) arm_cpu_features_string += " SWP";
+    if (arm_cpu_features.Half) arm_cpu_features_string += ", Half";
+    if (arm_cpu_features.Thumb) arm_cpu_features_string += ", Thumb";
+    if (arm_cpu_features.FastMult) arm_cpu_features_string += ", FastMult";
+    if (arm_cpu_features.VFP) arm_cpu_features_string += ", VFP";
+    if (arm_cpu_features.EDSP) arm_cpu_features_string += ", EDSP";
+    if (arm_cpu_features.ThumbEE) arm_cpu_features_string += ", ThumbEE";
+    if (arm_cpu_features.NEON) arm_cpu_features_string += ", NEON";
+    if (arm_cpu_features.VFPv3) arm_cpu_features_string += ", VFPv3";
+    if (arm_cpu_features.TLS) arm_cpu_features_string += ", TLS";
+    if (arm_cpu_features.VFPv4) arm_cpu_features_string += ", VFPv4";
+    if (arm_cpu_features.IDIVa) arm_cpu_features_string += ", IDIVa";
+    if (arm_cpu_features.IDIVt) arm_cpu_features_string += ", IDIVt";
+
+    DebugMessage(M64MSG_INFO, "%s", arm_cpu_features_string.c_str());
+}
+

jni/mupen64plus-core/src/r4300/new_dynarec/arm_cpu_features.h

@@ -0,0 +1,54 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - arm_cpu_features.h                                      *
+ *   Mupen64Plus homepage: http://code.google.com/p/mupen64plus/           *
+ *   Copyright (C) 2015 Gilles Siberlin                                    *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+#ifndef ARM_CPU_FEATURES_H
+#define ARM_CPU_FEATURES_H
+
+typedef struct
+{
+    unsigned char SWP;
+    unsigned char Half;
+    unsigned char Thumb;
+    unsigned char FastMult;
+    unsigned char VFP;
+    unsigned char EDSP;
+    unsigned char ThumbEE;
+    unsigned char NEON;
+    unsigned char VFPv3;
+    unsigned char TLS;
+    unsigned char VFPv4;
+    unsigned char IDIVa;
+    unsigned char IDIVt;
+}arm_cpu_features_t;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+extern arm_cpu_features_t arm_cpu_features;
+void detect_arm_cpu_features(void);
+void print_arm_cpu_features(void);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* ARM_CPU_FEATURES_H */

jni/mupen64plus-core/src/r4300/new_dynarec/assem_arm.c

@@ -2062,18 +2062,18 @@ static void emit_smull(u_int rs1, u_int rs2, u_int hi, u_int lo)
   output_w32(0xe0c00090|(hi<<16)|(lo<<12)|(rs2<<8)|rs1);
 }
 
-#ifdef HAVE_INTEGER_DIVIDE
 static void emit_sdiv(u_int rs1,u_int rs2,u_int rt)
 {
+  assert(arm_cpu_features.IDIVa);
   assem_debug("sdiv %s,%s,%s",regname[rt],regname[rs1],regname[rs2]);
   output_w32(0xe710f010|(rt<<16)|(rs2<<8)|rs1);
 }
 static void emit_udiv(u_int rs1,u_int rs2,u_int rt)
 {
+  assert(arm_cpu_features.IDIVa);
   assem_debug("udiv %s,%s,%s",regname[rt],regname[rs1],regname[rs2]);
   output_w32(0xe730f010|(rt<<16)|(rs2<<8)|rs1);
 }
-#endif
 
 static void emit_clz(int rs,int rt)
 {
@@ -4133,32 +4133,36 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs)
         signed char remainder=get_reg(i_regs->regmap,HIREG);
         assert(quotient>=0);
         assert(remainder>=0);
-#ifdef HAVE_INTEGER_DIVIDE
-        emit_test(d2,d2);
-        emit_jeq((int)out+16); // Division by zero
-        emit_sdiv(d1,d2,quotient);
-        emit_mul(quotient,d2,remainder);
-        emit_sub(d1,remainder,remainder);
-#else
-        emit_movs(d1,remainder);
-        emit_negmi(remainder,remainder);
-        emit_movs(d2,HOST_TEMPREG);
-        emit_jeq((int)out+52); // Division by zero
-        emit_negmi(HOST_TEMPREG,HOST_TEMPREG);
-        emit_clz(HOST_TEMPREG,quotient);
-        emit_shl(HOST_TEMPREG,quotient,HOST_TEMPREG);
-        emit_orimm(quotient,1<<31,quotient);
-        emit_shr(quotient,quotient,quotient);
-        emit_cmp(remainder,HOST_TEMPREG);
-        emit_subcs(remainder,HOST_TEMPREG,remainder);
-        emit_adcs(quotient,quotient,quotient);
-        emit_shrimm(HOST_TEMPREG,1,HOST_TEMPREG);
-        emit_jcc((int)out-16); // -4
-        emit_teq(d1,d2);
-        emit_negmi(quotient,quotient);
-        emit_test(d1,d1);
-        emit_negmi(remainder,remainder);
-#endif
+
+        if(arm_cpu_features.IDIVa)
+        {
+          emit_test(d2,d2);
+          emit_jeq((int)out+16); // Division by zero
+          emit_sdiv(d1,d2,quotient);
+          emit_mul(quotient,d2,remainder);
+          emit_sub(d1,remainder,remainder);
+        }
+        else
+        {
+          emit_movs(d1,remainder);
+          emit_negmi(remainder,remainder);
+          emit_movs(d2,HOST_TEMPREG);
+          emit_jeq((int)out+52); // Division by zero
+          emit_negmi(HOST_TEMPREG,HOST_TEMPREG);
+          emit_clz(HOST_TEMPREG,quotient);
+          emit_shl(HOST_TEMPREG,quotient,HOST_TEMPREG);
+          emit_orimm(quotient,1<<31,quotient);
+          emit_shr(quotient,quotient,quotient);
+          emit_cmp(remainder,HOST_TEMPREG);
+          emit_subcs(remainder,HOST_TEMPREG,remainder);
+          emit_adcs(quotient,quotient,quotient);
+          emit_shrimm(HOST_TEMPREG,1,HOST_TEMPREG);
+          emit_jcc((int)out-16); // -4
+          emit_teq(d1,d2);
+          emit_negmi(quotient,quotient);
+          emit_test(d1,d1);
+          emit_negmi(remainder,remainder);
+        }
       }
       if(opcode2[i]==0x1B) // DIVU
       {
@@ -4171,24 +4175,28 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs)
         assert(quotient>=0);
         assert(remainder>=0);
         emit_test(d2,d2);
-#ifdef HAVE_INTEGER_DIVIDE
-        emit_jeq((int)out+16); // Division by zero
-        emit_udiv(d1,d2,quotient);
-        emit_mul(quotient,d2,remainder);
-        emit_sub(d1,remainder,remainder);
-#else
-        emit_jeq((int)out+44); // Division by zero
-        emit_clz(d2,HOST_TEMPREG);
-        emit_movimm(1<<31,quotient);
-        emit_shl(d2,HOST_TEMPREG,d2);
-        emit_mov(d1,remainder);
-        emit_shr(quotient,HOST_TEMPREG,quotient);
-        emit_cmp(remainder,d2);
-        emit_subcs(remainder,d2,remainder);
-        emit_adcs(quotient,quotient,quotient);
-        emit_shrcc_imm(d2,1,d2);
-        emit_jcc((int)out-16); // -4
-#endif
+
+        if(arm_cpu_features.IDIVa)
+        {
+          emit_jeq((int)out+16); // Division by zero
+          emit_udiv(d1,d2,quotient);
+          emit_mul(quotient,d2,remainder);
+          emit_sub(d1,remainder,remainder);
+        }
+        else
+        {
+          emit_jeq((int)out+44); // Division by zero
+          emit_clz(d2,HOST_TEMPREG);
+          emit_movimm(1<<31,quotient);
+          emit_shl(d2,HOST_TEMPREG,d2);
+          emit_mov(d1,remainder);
+          emit_shr(quotient,HOST_TEMPREG,quotient);
+          emit_cmp(remainder,d2);
+          emit_subcs(remainder,d2,remainder);
+          emit_adcs(quotient,quotient,quotient);
+          emit_shrcc_imm(d2,1,d2);
+          emit_jcc((int)out-16); // -4
+        }
       }
     }
     else // 64-bit
@@ -4568,6 +4576,10 @@ static void do_clear_cache()
 
 // CPU-architecture-specific initialization
 static void arch_init() {
+
+  detect_arm_cpu_features();
+  print_arm_cpu_features();
+
   rounding_modes[0]=0x0<<22; // round
   rounding_modes[1]=0x3<<22; // trunc
   rounding_modes[2]=0x1<<22; // ceil

jni/mupen64plus-core/src/r4300/new_dynarec/assem_arm.h

@@ -13,7 +13,6 @@
 //#define REG_PREFETCH 1
 #define HAVE_CONDITIONAL_CALL 1
 #define RAM_OFFSET 1
-#define HAVE_INTEGER_DIVIDE 1
 
 /* ARM calling convention:
    r0-r3, r12: caller-save

jni/mupen64plus-core/src/r4300/new_dynarec/new_dynarec.c

@@ -44,6 +44,7 @@
 #if NEW_DYNAREC == NEW_DYNAREC_X86
 #include "assem_x86.h"
 #elif NEW_DYNAREC == NEW_DYNAREC_ARM
+#include "arm_cpu_features.h"
 #include "assem_arm.h"
 #else
 #error Unsupported dynarec architecture