[x264-devel] [PATCH 3/3] checkasm: arm: Check register clobbering
Martin Storsjö
martin at martin.st
Thu Aug 27 23:15:03 CEST 2015
Use a separate function prototype which returns uint64_t instead
of the normal one that returns intptr_t.
Use two separate sets of functions, depending on whether neon
is available.
---
Instead of using a separate checkasm_call function for a 64 bit
return value, one could also just declare the second one with
asm("x264_checkasm_call_neon") to have it map to the same symbol,
but I guess it's easier to duplicate than to deal with the
potential portability issues.
The same would have to be done for x86 32 bit as well, if the
sa8d_satd function would be implemented there.
---
Makefile | 1 +
tools/checkasm-arm.S | 180 ++++++++++++++++++++++++++++++++++++++++++++++++++
tools/checkasm.c | 25 ++++++-
3 files changed, 204 insertions(+), 2 deletions(-)
create mode 100644 tools/checkasm-arm.S
diff --git a/Makefile b/Makefile
index 4feef33..d0b1633 100644
--- a/Makefile
+++ b/Makefile
@@ -122,6 +122,7 @@ ASMSRC += common/arm/cpu-a.S common/arm/pixel-a.S common/arm/mc-a.S \
common/arm/predict-a.S common/arm/bitstream-a.S
SRCS += common/arm/mc-c.c common/arm/predict-c.c
OBJASM = $(ASMSRC:%.S=%.o)
+OBJCHK += tools/checkasm-arm.o
endif
endif
diff --git a/tools/checkasm-arm.S b/tools/checkasm-arm.S
new file mode 100644
index 0000000..8b3baac
--- /dev/null
+++ b/tools/checkasm-arm.S
@@ -0,0 +1,180 @@
+/****************************************************************************
+ * checkasm-arm.S: assembly check tool
+ *****************************************************************************
+ * Copyright (C) 2015 x264 project
+ *
+ * Authors: Martin Storsjo <martin at martin.st>
+ *
+ * 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 02111, USA.
+ *
+ * This program is also available under a commercial proprietary license.
+ * For more information, contact us at licensing at x264.com.
+ *****************************************************************************/
+
+#include "../common/arm/asm.S"
+
+.section .rodata
+.align 4
+register_init:
+.quad 0x21f86d66c8ca00ce
+.quad 0x75b6ba21077c48ad
+.quad 0xed56bb2dcb3c7736
+.quad 0x8bda43d3fd1a7e06
+.quad 0xb64a9c9e5d318408
+.quad 0xdf9a54b303f1d3a3
+.quad 0x4a75479abd64e097
+.quad 0x249214109d5d1c88
+
+error_message:
+.asciz "failed to preserve register"
+
+.text
+
+@ max number of args used by any x264 asm function.
+#define MAX_ARGS 15
+
+#define ARG_STACK 4*(MAX_ARGS - 2)
+#define PUSHED 16*4 + 4*10
+
+function x264_checkasm_call_neon
+ push {r4-r11, lr}
+ vpush {q4-q7}
+
+ movrel r12, register_init
+ vldm r12, {q4-q7}
+ ldm r12, {r4-r11}
+
+ push {r1}
+
+ sub sp, sp, #ARG_STACK
+.equ pos, 0
+.rept MAX_ARGS-2
+ ldr r12, [sp, #ARG_STACK + PUSHED + 8 + pos]
+ str r12, [sp, #pos]
+.equ pos, pos + 4
+.endr
+
+ mov r12, r0
+ mov r0, r2
+ mov r1, r3
+ ldrd r2, r3, [sp, #ARG_STACK + PUSHED]
+ blx r12
+ add sp, sp, #ARG_STACK
+ pop {r2}
+
+ push {r0, r1}
+ movrel r12, register_init
+ vldm r12, {q0-q3}
+ veor q0, q0, q4
+ veor q1, q1, q5
+ veor q2, q2, q6
+ veor q3, q3, q7
+ vorr q0, q0, q1
+ vorr q0, q0, q2
+ vorr q0, q0, q3
+ vorr d0, d0, d1
+ vrev64.32 d1, d0
+ vorr d0, d0, d1
+ vmov.32 r3, d0[0]
+
+.macro check_reg reg1, reg2
+ ldrd r0, r1, [r12], #8
+ eor r0, r0, \reg1
+ eor r1, r1, \reg2
+ orr r3, r3, r0
+ orr r3, r3, r1
+.endm
+ check_reg r4, r5
+ check_reg r6, r7
+ check_reg r8, r9
+ check_reg r10, r11
+.purgem check_reg
+
+ cmp r3, #0
+ beq 0f
+
+ mov r12, #0
+ str r12, [r2]
+ movrel r0, error_message
+ bl puts
+0:
+ pop {r0, r1}
+ vpop {q4-q7}
+ pop {r4-r11, pc}
+endfunc
+
+function x264_checkasm_call_neon64
+ b X(x264_checkasm_call_neon)
+endfunc
+
+#undef PUSHED
+#define PUSHED 4*10
+
+function x264_checkasm_call_noneon
+ push {r4-r11, lr}
+
+ movrel r12, register_init
+ ldm r12, {r4-r11}
+
+ push {r1}
+
+ sub sp, sp, #ARG_STACK
+.equ pos, 0
+.rept MAX_ARGS-2
+ ldr r12, [sp, #ARG_STACK + PUSHED + 8 + pos]
+ str r12, [sp, #pos]
+.equ pos, pos + 4
+.endr
+
+ mov r12, r0
+ mov r0, r2
+ mov r1, r3
+ ldrd r2, r3, [sp, #ARG_STACK + PUSHED]
+ blx r12
+ add sp, sp, #ARG_STACK
+ pop {r2}
+
+ push {r0, r1}
+ movrel r12, register_init
+ mov r3, #0
+
+.macro check_reg reg1, reg2
+ ldrd r0, r1, [r12], #8
+ eor r0, r0, \reg1
+ eor r1, r1, \reg2
+ orr r3, r3, r0
+ orr r3, r3, r1
+.endm
+ check_reg r4, r5
+ check_reg r6, r7
+ check_reg r8, r9
+ check_reg r10, r11
+.purgem check_reg
+
+ cmp r3, #0
+ beq 0f
+
+ mov r12, #0
+ str r12, [r2]
+ movrel r0, error_message
+ bl puts
+0:
+ pop {r0, r1}
+ pop {r4-r11, pc}
+endfunc
+
+function x264_checkasm_call_noneon64
+ b X(x264_checkasm_call_noneon)
+endfunc
diff --git a/tools/checkasm.c b/tools/checkasm.c
index 01a97c9..ca345f1 100644
--- a/tools/checkasm.c
+++ b/tools/checkasm.c
@@ -231,6 +231,15 @@ intptr_t x264_checkasm_call( intptr_t (*func)(), int *ok, ... );
intptr_t x264_checkasm_call( intptr_t (*func)(), int *ok, ... );
#endif
+#if ARCH_ARM
+intptr_t x264_checkasm_call_neon( intptr_t (*func)(), int *ok, ... );
+uint64_t x264_checkasm_call_neon64( intptr_t (*func)(), int *ok, ... );
+intptr_t x264_checkasm_call_noneon( intptr_t (*func)(), int *ok, ... );
+uint64_t x264_checkasm_call_noneon64( intptr_t (*func)(), int *ok, ... );
+intptr_t (*x264_checkasm_call)( intptr_t (*func)(), int *ok, ... ) = x264_checkasm_call_noneon;
+uint64_t (*x264_checkasm_call64)( intptr_t (*func)(), int *ok, ... ) = x264_checkasm_call_noneon64;
+#endif
+
#define call_c1(func,...) func(__VA_ARGS__)
#if ARCH_X86_64
@@ -248,12 +257,18 @@ void x264_checkasm_stack_clobber( uint64_t clobber, ... );
uint64_t r = (rand() & 0xffff) * 0x0001000100010001ULL; \
x264_checkasm_stack_clobber( r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r ); /* max_args+6 */ \
x264_checkasm_call(( intptr_t(*)())func, &ok, 0, 0, 0, 0, __VA_ARGS__ ); })
-#elif ARCH_X86 || (ARCH_AARCH64 && !defined(__APPLE__))
+#elif ARCH_X86 || (ARCH_AARCH64 && !defined(__APPLE__)) || ARCH_ARM
#define call_a1(func,...) x264_checkasm_call( (intptr_t(*)())func, &ok, __VA_ARGS__ )
#else
#define call_a1 call_c1
#endif
+#if ARCH_ARM
+#define call_a1_64(func,...) x264_checkasm_call64( (intptr_t(*)())func, &ok, __VA_ARGS__ )
+#else
+#define call_a1_64 call_a1
+#endif
+
#define call_bench(func,cpu,...)\
if( do_bench && !strncmp(func_name, bench_pattern, bench_pattern_len) )\
{\
@@ -286,6 +301,7 @@ void x264_checkasm_stack_clobber( uint64_t clobber, ... );
#define call_c(func,...) ({ call_c2(func,__VA_ARGS__); call_c1(func,__VA_ARGS__); })
#define call_a2(func,...) ({ call_bench(func,cpu_new,__VA_ARGS__); })
#define call_c2(func,...) ({ call_bench(func,0,__VA_ARGS__); })
+#define call_a64(func,...) ({ call_a2(func,__VA_ARGS__); call_a1_64(func,__VA_ARGS__); })
static int check_pixel( int cpu_ref, int cpu_new )
@@ -372,7 +388,7 @@ static int check_pixel( int cpu_ref, int cpu_new )
{
uint32_t cost8_c = pixel_c.sa8d[PIXEL_16x16]( pbuf1, 16, pbuf2, 64 );
uint32_t cost4_c = pixel_c.satd[PIXEL_16x16]( pbuf1, 16, pbuf2, 64 );
- uint64_t res_a = call_a( pixel_asm.sa8d_satd[PIXEL_16x16], pbuf1, (intptr_t)16, pbuf2, (intptr_t)64 );
+ uint64_t res_a = call_a64( pixel_asm.sa8d_satd[PIXEL_16x16], pbuf1, (intptr_t)16, pbuf2, (intptr_t)64 );
uint32_t cost8_a = res_a;
uint32_t cost4_a = res_a >> 32;
if( cost8_a != cost8_c || cost4_a != cost4_c )
@@ -2786,6 +2802,11 @@ static int check_all_flags( void )
ret = check_all_funcs( 0, X264_CPU_ALTIVEC );
}
#elif ARCH_ARM
+ if( cpu_detect & X264_CPU_NEON )
+ {
+ x264_checkasm_call = x264_checkasm_call_neon;
+ x264_checkasm_call64 = x264_checkasm_call_neon64;
+ }
if( cpu_detect & X264_CPU_ARMV6 )
ret |= add_flags( &cpu0, &cpu1, X264_CPU_ARMV6, "ARMv6" );
if( cpu_detect & X264_CPU_NEON )
--
1.7.10.4
More information about the x264-devel
mailing list