VFP Applications

• Automotive control applications
  • Powertrain
  • ABS, Traction control & active suspension
• 3D Graphics
  • Digital consumer products
  • Set-top boxes, games consoles
• Imaging
  • Laser printers, still digital cameras, digital video cameras
• Industrial control systems
  • Motion controls

4.2  VFP的架构版本

1. VFPv1 is obsolete. Details are available on request from ARM.
2. VFPv2 is an optional extension to the ARM instruction set in the ARMv5TE, ARMv5TEJ and ARMv6 architectures.
3. VFPv3 is an optional extension to the ARM, Thumb® and ThumbEE instruction sets in the ARMv7-A and ARMv7-R profiles. VFPv3 implementation is with either thirty-two or sixteen double word registers. The terms VFPv3-D32 and VFPv3-D16 distinguish between these two implementation options. Extending VFPv3 uses the half-precision extensions that provide conversion functions in both directions between half-precision floating-point and single-precision floating-point.

1      浮点类型-mfloat-abi

1.1      选项

-mfloat-abi=soft/softfp/hard
支持3种类型，各类型含义如下：
 soft
不使用硬件浮点单元，gcc使用软浮点库来完成浮点运算。适用于不含FPU的CPU。
 softfp
使用硬浮点单元，会生成硬浮点指令，生成何种类型的硬浮点指令由-mfpu选项指定。调用接口的规则和soft选项一致。
 hard
使用硬浮点单元，生成硬浮点指令。与softfp的区别在于调用接口的规则不同。

1.2  示例

1.2.1 示例代码
float mul(float a, float b)
{
return a*b;
}
1.2.2 soft选项
编译及反汇编命令
arm-linux-gcc -Wall -march=armv7-a -mcpu=cortex-a9 -mfloat-abi=soft -c test.c
arm-linux-objdump -D test.o | less
生成的汇编指令
00000000 <mul>:
0: e92d4800 push {fp, lr}
4: e28db004 add fp, sp, #4
8: e24dd008 sub sp, sp, #8
c: e50b0008 str r0, [fp, #-8]
10: e50b100c str r1, [fp, #-12]
14: e51b0008 ldr r0, [fp, #-8]
18: e51b100c ldr r1, [fp, #-12]
1c: ebfffffe bl 0 <__aeabi_fmul>
20: e1a03000 mov r3, r0
24: e1a00003 mov r0, r3
28: e24bd004 sub sp, fp, #4
2c: e8bd8800 pop {fp, pc}
可以看出是调用__aeabi_fmul接口来进行浮点运算。
1.2.3 softfp选项
编译命令
arm-linux-gcc -Wall -march=armv7-a -mcpu=cortex-a9 -mfloat-abi=softfp -mfpu=vfpv3-d16 -c test.c
生成的汇编指令
00000000 <mul>:
0: e52db004 push {fp} ; (str fp, [sp, #-4]!)
4: e28db000 add fp, sp, #0
8: e24dd00c sub sp, sp, #12
c: e50b0008 str r0, [fp, #-8]
10: e50b100c str r1, [fp, #-12]
14: ed1b7a02 vldr s14, [fp, #-8]
18: ed5b7a03 vldr s15, [fp, #-12]
1c: ee677a27 vmul.f32 s15, s14, s15
20: ee173a90 vmov r3, s15
24: e1a00003 mov r0, r3
28: e28bd000 add sp, fp, #0
2c: e49db004 pop {fp} ; (ldr fp, [sp], #4)
30: e12fff1e bx lr
生成了vxxx的硬浮点指令。并且可以看出和soft一样，都是用r0,r1来传递形参。
1.2.4 hard
编译命令
arm-linux-gcc -Wall -march=armv7-a -mcpu=cortex-a9 -mfloat-abi=hard -mfpu=vfpv3-d16 -c test.c
生成的汇编指令
00000000 <mul>:
0: e52db004 push {fp} ; (str fp, [sp, #-4]!)
4: e28db000 add fp, sp, #0
8: e24dd00c sub sp, sp, #12
c: ed0b0a02 vstr s0, [fp, #-8]
10: ed4b0a03 vstr s1, [fp, #-12]
14: ed1b7a02 vldr s14, [fp, #-8]
18: ed5b7a03 vldr s15, [fp, #-12]
1c: ee677a27 vmul.f32 s15, s14, s15
20: eeb00a67 vmov.f32 s0, s15
24: e28bd000 add sp, fp, #0
28: e49db004 pop {fp} ; (ldr fp, [sp], #4)
2c: e12fff1e bx lr
同样生成了硬浮点指令，与softfp的区别在于，这里使用FPU的寄存器s0、s1来传递形参。

2 使用NEON

2.1 选项

-O3 -mfloat-abi=softfp -mfpu=neon -ftree-vectorize
neon可以做浮点运算，有了neon，可以不使用vfp。
为了提升生成的代码性能，应该使用neon intrinsics的方式来写代码。
2.2 示例
普通代码
void NeonTest(int * x, int * y, int * z)
{
int i;
for(i=0;i<200;i++) {
z[i] = x[i] + y[i];
}
}
neon intrinsics格式的代码
#include <arm_neon.h>
void intrinsics(uint32_t *x, uint32_t *y, uint32_t *z)
{
int i;
uint32x4_t x4,y4; // These 128 bit registers will contain 4 values from the x array and 4 values from the y array
uint32x4_t z4; // This 128 bit register will contain the 4 results from the add intrinsic
uint32_t *ptra = x; // pointer to the x array data
uint32_t *ptrb = y; // pointer to the y array data
uint32_t *ptrz = z; // pointer to the z array data
for(i=0; i < 200/4; i++)
{
x4 = vld1q_u32(ptra); // intrinsic to load x4 with 4 values from x
y4 = vld1q_u32(ptrb); // intrinsic to load y4
z4=vaddq_u32(x4,y4); // intrinsic to add z4=x4+y4
vst1q_u32(ptrz, z4); // store the 4 results to z
ptra+=4; // increment pointers
ptrb+=4;
ptrz+=4;
}
}