<rdar://problem/12512722> Use arm divide instruction if available

git-svn-id: https://llvm.org/svn/llvm-project/compiler-rt/trunk@182665 91177308-0d34-0410-b5e6-96231b3b80d8

Author: Nick Kledzik

lib/arm/divmodsi4.S

@@ -24,6 +24,18 @@
 .syntax unified
 .align 3
 DEFINE_COMPILERRT_FUNCTION(__divmodsi4)
+#if __ARM_ARCH_7S__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divzero)
+	mov 	r3, r0
+	sdiv	r0, r3, r1
+	mls 	r1, r0, r1, r3
+	str 	r1, [r2]
+	bx  	lr
+LOCAL_LABEL(divzero):
+	mov     r0, #0
+	bx      lr
+#else
     ESTABLISH_FRAME
 //  Set aside the sign of the quotient and modulus, and the address for the
 //  modulus.
@@ -45,3 +57,4 @@ DEFINE_COMPILERRT_FUNCTION(__divmodsi4)
     sub     r1,     r1, r5, asr #31
     str     r1,    [r6]
     CLEAR_FRAME_AND_RETURN
+#endif

lib/arm/modsi3.S

@@ -23,6 +23,16 @@
 .syntax unified
 .align 3
 DEFINE_COMPILERRT_FUNCTION(__modsi3)
+#if __ARM_ARCH_7S__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divzero)
+	sdiv	r2, r0, r1
+	mls 	r0, r2, r1, r0
+	bx      lr
+LOCAL_LABEL(divzero):
+	mov     r0, #0
+	bx      lr
+#else
     ESTABLISH_FRAME
     //  Set aside the sign of the dividend.
     mov     r4,     r0
@@ -37,3 +47,4 @@ DEFINE_COMPILERRT_FUNCTION(__modsi3)
     eor     r0,     r0, r4, asr #31
     sub     r0,     r0, r4, asr #31
     CLEAR_FRAME_AND_RETURN
+#endif

lib/arm/udivmodsi4.S

@@ -31,6 +31,18 @@
 .syntax unified
 .align 3
 DEFINE_COMPILERRT_FUNCTION(__udivmodsi4)
+#if __ARM_ARCH_7S__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divzero)
+	mov 	r3, r0
+	udiv	r0, r3, r1
+	mls 	r1, r0, r1, r3
+	str 	r1, [r2]
+	bx  	lr
+LOCAL_LABEL(divzero):
+	mov     r0, #0
+	bx      lr
+#else
 //  We use a simple digit by digit algorithm; before we get into the actual 
 //  divide loop, we must calculate the left-shift amount necessary to align
 //  the MSB of the divisor with that of the dividend (If this shift is
@@ -78,3 +90,4 @@ LOCAL_LABEL(return):
     str     a,     [r2]
     mov     r0,     q
     CLEAR_FRAME_AND_RETURN
+#endif

lib/arm/umodsi3.S

@@ -23,6 +23,16 @@
 .syntax unified
 .align 3
 DEFINE_COMPILERRT_FUNCTION(__umodsi3)
+#if __ARM_ARCH_7S__
+	tst     r1, r1
+	beq     LOCAL_LABEL(divzero)
+	udiv	r2, r0, r1
+	mls     r0, r2, r1, r0
+	bx      lr
+LOCAL_LABEL(divzero):
+	mov     r0, #0
+	bx      lr
+#else
 //  We use a simple digit by digit algorithm; before we get into the actual 
 //  divide loop, we must calculate the left-shift amount necessary to align
 //  the MSB of the divisor with that of the dividend.
@@ -56,3 +66,4 @@ LOCAL_LABEL(mainLoop):
     subs    r,      a,  b
     movhs   a,      r
     bx      lr
+#endif

test/timing/modsi3.c

@@ -0,0 +1,52 @@
+#include "timing.h"
+#include <stdio.h>
+
+#define INPUT_TYPE int32_t
+#define INPUT_SIZE 256
+#define FUNCTION_NAME __modsi3
+
+#ifndef LIBNAME
+#define LIBNAME UNKNOWN
+#endif
+
+#define LIBSTRING		LIBSTRINGX(LIBNAME)
+#define LIBSTRINGX(a)	LIBSTRINGXX(a)
+#define LIBSTRINGXX(a)	#a
+
+INPUT_TYPE FUNCTION_NAME(INPUT_TYPE input1, INPUT_TYPE input2);
+
+int main(int argc, char *argv[]) {
+	INPUT_TYPE input1[INPUT_SIZE];
+	INPUT_TYPE input2[INPUT_SIZE];
+	int i, j;
+	
+	srand(42);
+	
+	// Initialize the input array with data of various sizes.
+	for (i=0; i<INPUT_SIZE; ++i) {
+		input1[i] = rand();
+		input2[i] = rand() + 1;
+	}
+	
+	int64_t fixedInput = INT64_C(0x1234567890ABCDEF);
+	
+	double bestTime = __builtin_inf();
+	void *dummyp;
+	for (j=0; j<1024; ++j) {
+		
+		uint64_t startTime = mach_absolute_time();
+		for (i=0; i<INPUT_SIZE; ++i)
+			FUNCTION_NAME(input1[i], input2[i]);
+		uint64_t endTime = mach_absolute_time();
+		
+		double thisTime = intervalInCycles(startTime, endTime);
+		bestTime = __builtin_fmin(thisTime, bestTime);
+		
+		// Move the stack alignment between trials to eliminate (mostly) aliasing effects
+		dummyp = alloca(1);
+	}
+	
+	printf("%16s: %f cycles.\n", LIBSTRING, bestTime / (double) INPUT_SIZE);
+	
+	return 0;
+}