View
1.559
Download
1
Category
Tags:
Preview:
DESCRIPTION
The iPhone has a surprisingly powerful engine under that shiny hood when it comes to floating-point computations. This is something that surprises a lot of programmers because by default, things can slow down a lot whenever any floating point numbers are involved. This session will explain the secrets to unlocking maximum performance for floating point calculations, from the mysteries of Thumb mode, to harnessing the full power of the forgotten vector floating point unit. Stay away from this session if he thought of reading or even (gasp!) writing assembly code scares you.
Citation preview
Cranking Floating Point Performance Up To 11
Noel LlopisSnappy Touch
http://twitter.com/snappytouchnoel@snappytouch.com
http://gamesfromwithin.com
Floating Point Performance
Floating point numbers
Floating point numbers
• Representation of rational numbers
Floating point numbers
• Representation of rational numbers
• 1.2345, -0.8374, 2.0000, 14388439.34, etc
Floating point numbers
• Representation of rational numbers
• 1.2345, -0.8374, 2.0000, 14388439.34, etc
• Following IEEE 754 format
Floating point numbers
• Representation of rational numbers
• 1.2345, -0.8374, 2.0000, 14388439.34, etc
• Following IEEE 754 format
• Single precision: 32 bits
Floating point numbers
• Representation of rational numbers
• 1.2345, -0.8374, 2.0000, 14388439.34, etc
• Following IEEE 754 format
• Single precision: 32 bits
• Double precision: 64 bits
Floating point numbers
Floating point numbers
Why floating point performance?
Why floating point performance?
• Most games use floating point numbers for most of their calculations
Why floating point performance?
• Most games use floating point numbers for most of their calculations
• Positions, velocities, physics, etc, etc.
Why floating point performance?
• Most games use floating point numbers for most of their calculations
• Positions, velocities, physics, etc, etc.
• Maybe not so much for regular apps
CPU
CPU
• 32-bit RISC ARM 11
CPU
• 32-bit RISC ARM 11
• 400-535Mhz
CPU
• 32-bit RISC ARM 11
• 400-535Mhz
• iPhone 2G/3G and iPod Touch 1st and 2nd gen
CPU (iPhone 3GS)
CPU (iPhone 3GS)
• Cortex-A8 600MHz
CPU (iPhone 3GS)
• Cortex-A8 600MHz
• More advanced architecture
CPU
CPU
• No floating point support in the ARM CPU!!!
How about integer math?
How about integer math?
• No need to do any floating point operations
How about integer math?
• No need to do any floating point operations
• Fully supported in the ARM processor
How about integer math?
• No need to do any floating point operations
• Fully supported in the ARM processor
• But...
Integer Divide
Integer Divide
Integer Divide
There is no integer divide
Fixed-point arithmetic
Fixed-point arithmetic
• Sometimes integer arithmetic doesn’t cut it
Fixed-point arithmetic
• Sometimes integer arithmetic doesn’t cut it
• You need to represent rational numbers
Fixed-point arithmetic
• Sometimes integer arithmetic doesn’t cut it
• You need to represent rational numbers
• Can use a fixed-point library.
Fixed-point arithmetic
• Sometimes integer arithmetic doesn’t cut it
• You need to represent rational numbers
• Can use a fixed-point library.
• Performs rational arithmetic with integer values at a reduced range/resolution.
Fixed-point arithmetic
• Sometimes integer arithmetic doesn’t cut it
• You need to represent rational numbers
• Can use a fixed-point library.
• Performs rational arithmetic with integer values at a reduced range/resolution.
• Not so great...
Floating point support
Floating point support
• There’s a floating point unit
Floating point support
• There’s a floating point unit
• Compiled C/C++/ObjC code uses the VFP unit for any floating point operations.
Sample program
Sample program struct Particle { float x, y, z; float vx, vy, vz; };
Sample program struct Particle { float x, y, z; float vx, vy, vz; };
for (int i=0; i<MaxParticles; ++i){ Particle& p = s_particles[i]; p.x += p.vx*dt; p.y += p.vy*dt; p.z += p.vz*dt; p.vx *= drag; p.vy *= drag; p.vz *= drag;}
Sample program struct Particle { float x, y, z; float vx, vy, vz; };
for (int i=0; i<MaxParticles; ++i){ Particle& p = s_particles[i]; p.x += p.vx*dt; p.y += p.vy*dt; p.z += p.vz*dt; p.vx *= drag; p.vy *= drag; p.vz *= drag;}
• 7.2 seconds on an iPod Touch 2nd gen
Floating point support
Floating point support
Trust no one!
Floating point support
Trust no one!When in doubt, check the
assembly generated
Floating point support
Thumb Mode
Thumb Mode
Thumb Mode• CPU has a special thumb
mode.
Thumb Mode• CPU has a special thumb
mode.
• Less memory, maybe better performance.
Thumb Mode• CPU has a special thumb
mode.
• Less memory, maybe better performance.
• No floating point support.
Thumb Mode• CPU has a special thumb
mode.
• Less memory, maybe better performance.
• No floating point support.
• Every time there’s an fp operation, it switches out of Thumb, does the fp operation, and switches back on.
Thumb Mode
Thumb Mode
• It’s on by default!
Thumb Mode
• It’s on by default!
• Potentially HUGE wins turning it off.
Thumb Mode
• It’s on by default!
• Potentially HUGE wins turning it off.
Thumb Mode
Thumb Mode
• Turning off Thumb mode increased performance in Flower Garden by over 2x
Thumb Mode
• Turning off Thumb mode increased performance in Flower Garden by over 2x
• Heavy usage of floating point operations though
Thumb Mode
• Turning off Thumb mode increased performance in Flower Garden by over 2x
• Heavy usage of floating point operations though
• Most games will probably benefit from turning it off (especially 3D games)
2.6 seconds!
ARM assemblyDISCLAIMER:
ARM assembly
I’m not an ARM assembly expert!!!DISCLAIMER:
ARM assembly
I’m not an ARM assembly expert!!!DISCLAIMER:
ARM assembly
I’m not an ARM assembly expert!!!DISCLAIMER:
ARM assembly
I’m not an ARM assembly expert!!!DISCLAIMER:
Z80!!!
ARM assembly
ARM assembly
• Hit the docs
ARM assembly
• Hit the docs
• References included in your USB card
ARM assembly
• Hit the docs
• References included in your USB card
• Or download them from the ARM site
ARM assembly
• Hit the docs
• References included in your USB card
• Or download them from the ARM site
• http://bit.ly/arminfo
ARM assembly
ARM assembly
• Reading assembly is a very important skill for high-performance programming
ARM assembly
• Reading assembly is a very important skill for high-performance programming
• Writing is more specialized. Most people don’t need to.
VFP unit
VFP unitA0
VFP unitA0
+
VFP unitA0
B0+
VFP unitA0
B0+
=
VFP unitA0
B0+
C0=
VFP unitA0
B0+
C0=
A1
B1+
C1=
VFP unitA0
B0+
C0=
A1
B1+
C1=
A2
B2+
C2=
VFP unitA0
B0+
C0=
A1
B1+
C1=
A2
B2+
C2=
A3
B3+
C3=
VFP unit
VFP unitA0 A1 A2 A3
VFP unit
+A0 A1 A2 A3
VFP unit
+A0 A1 A2 A3
B0 B1 B2 B3
VFP unit
+
=
A0 A1 A2 A3
B0 B1 B2 B3
VFP unit
+
=
A0 A1 A2 A3
B0 B1 B2 B3
C0 C1 C2 C3
VFP unit
+
=
A0 A1 A2 A3
B0 B1 B2 B3
C0 C1 C2 C3
Sweet! How do we use the vfp?
"fldmias %2, {s8-s23} \n\t" "fldmias %1!, {s0-s3} \n\t" "fmuls s24, s8, s0 \n\t" "fmacs s24, s12, s1 \n\t"
"fldmias %1!, {s4-s7} \n\t"
"fmacs s24, s16, s2 \n\t" "fmacs s24, s20, s3 \n\t" "fstmias %0!, {s24-s27} \n\t"
Like this!
Writing vfp assembly
Writing vfp assembly
• There are two parts to it
Writing vfp assembly
• There are two parts to it
• How to write any assembly in gcc
Writing vfp assembly
• There are two parts to it
• How to write any assembly in gcc
• Learning ARM and VPM assembly
vfpmath library
vfpmath library
• Already done a lot of work for you
vfpmath library
• Already done a lot of work for you
• http://code.google.com/p/vfpmathlibrary
vfpmath library
• Already done a lot of work for you
• http://code.google.com/p/vfpmathlibrary
• Vector/matrix math
vfpmath library
• Already done a lot of work for you
• http://code.google.com/p/vfpmathlibrary
• Vector/matrix math
• Might not be exactly what you need, but it’s a great starting point
Assembly in gcc
• Only use it when targeting the device
Assembly in gcc
• Only use it when targeting the device
#include <TargetConditionals.h>#if (TARGET_IPHONE_SIMULATOR == 0) && (TARGET_OS_IPHONE == 1) #define USE_VFP#endif
Assembly in gcc
• The basics
asm (“cmp r2, r1”);
Assembly in gcc
• The basics
asm (“cmp r2, r1”);
http://www.ibiblio.org/gferg/ldp/GCC-Inline-Assembly-HOWTO.html
Assembly in gcc
• Multiple lines
asm ( “mov r0, #1000\n\t” “cmp r2, r1\n\t”);
Assembly in gcc• Accessing C variables
asm (//assembly code : // output operands : // input operands : // clobbered registers);
Assembly in gcc• Accessing C variables
asm (//assembly code : // output operands : // input operands : // clobbered registers);
int src = 19; int dest = 0; asm volatile ( "add %0, %1, #42" : "=r" (dest) : "r" (src) : );
Assembly in gcc• Accessing C variables
asm (//assembly code : // output operands : // input operands : // clobbered registers);
int src = 19; int dest = 0; asm volatile ( "add %0, %1, #42" : "=r" (dest) : "r" (src) : );
%0, %1, etc are the variables in order
Assembly in gcc
Assembly in gcc int src = 19; int dest = 0; asm volatile ( "add r10, %1, #42\n\t" "add %0, r10, #33\n\t" : "=r" (dest) : "r" (src) : "r10" );
Assembly in gcc int src = 19; int dest = 0; asm volatile ( "add r10, %1, #42\n\t" "add %0, r10, #33\n\t" : "=r" (dest) : "r" (src) : "r10" );
Clobber register list are registers used by
the asm block
Assembly in gcc int src = 19; int dest = 0; asm volatile ( "add r10, %1, #42\n\t" "add %0, r10, #33\n\t" : "=r" (dest) : "r" (src) : "r10" );
Clobber register list are registers used by
the asm block
volatile prevents “optimizations”
VFP asmFour banks of 8 32-bit registers each
VFP asmFour banks of 8 32-bit registers each
#define VFP_VECTOR_LENGTH(VEC_LENGTH) "fmrx r0, fpscr \n\t" \ "bic r0, r0, #0x00370000 \n\t" \ "orr r0, r0, #0x000" #VEC_LENGTH "0000 \n\t" \ "fmxr fpscr, r0 \n\t"
VFP asm
VFP asm
VFP asmfor (int i=0; i<MaxParticles; ++i){ Particle& p = s_particles[i]; p.x += p.vx*dt; p.y += p.vy*dt; p.z += p.vz*dt; p.vx *= drag; p.vy *= drag; p.vz *= drag;}
VFP asm for (int i=0; i<MaxParticles; ++i) { Particle* p = &s_particles[i]; asm volatile ( "fldmias %0, {s0-s5} \n\t" "fldmias %1, {s6-s8} \n\t" "fldmias %2, {s9-s11} \n\t" "fmacs s0, s3, s6 \n\t" "fmuls s3, s3, s9 \n\t" "fstmias %0, {s0-s5} \n\t" : "=r" (p) : "r" (p), "r" (dtArray), "r" (dragArray) : ); }
for (int i=0; i<MaxParticles; ++i){ Particle& p = s_particles[i]; p.x += p.vx*dt; p.y += p.vy*dt; p.z += p.vz*dt; p.vx *= drag; p.vy *= drag; p.vz *= drag;}
VFP asm for (int i=0; i<MaxParticles; ++i) { Particle* p = &s_particles[i]; asm volatile ( "fldmias %0, {s0-s5} \n\t" "fldmias %1, {s6-s8} \n\t" "fldmias %2, {s9-s11} \n\t" "fmacs s0, s3, s6 \n\t" "fmuls s3, s3, s9 \n\t" "fstmias %0, {s0-s5} \n\t" : "=r" (p) : "r" (p), "r" (dtArray), "r" (dragArray) : ); }
for (int i=0; i<MaxParticles; ++i){ Particle& p = s_particles[i]; p.x += p.vx*dt; p.y += p.vy*dt; p.z += p.vz*dt; p.vx *= drag; p.vy *= drag; p.vz *= drag;}
Was: 2.6 seconds
VFP asm for (int i=0; i<MaxParticles; ++i) { Particle* p = &s_particles[i]; asm volatile ( "fldmias %0, {s0-s5} \n\t" "fldmias %1, {s6-s8} \n\t" "fldmias %2, {s9-s11} \n\t" "fmacs s0, s3, s6 \n\t" "fmuls s3, s3, s9 \n\t" "fstmias %0, {s0-s5} \n\t" : "=r" (p) : "r" (p), "r" (dtArray), "r" (dragArray) : ); }
for (int i=0; i<MaxParticles; ++i){ Particle& p = s_particles[i]; p.x += p.vx*dt; p.y += p.vy*dt; p.z += p.vz*dt; p.vx *= drag; p.vy *= drag; p.vz *= drag;}
Was: 2.6 secondsNow: 1.4 seconds!!
VFP asmLet’s do 6 operations at once!
struct Particle2 { float x0, y0, z0; float x1, y1, z1; float vx0, vy0, vz0; float vx1, vy1, vz1; };
VFP asm for (int i=0; i<iterations; ++i) { Particle2* p = &s_particles2[i]; asm volatile ( "fldmias %0, {s0-s11} \n\t" "fldmias %1, {s12-s17} \n\t" "fldmias %2, {s18-s23} \n\t" "fmacs s0, s6, s12 \n\t" "fmuls s6, s6, s18 \n\t" "fstmias %0, {s0-s11} \n\t" : "=r" (p) : "r" (p), "r" (dtArray), "r" (dragArray) : ); }
VFP asm for (int i=0; i<iterations; ++i) { Particle2* p = &s_particles2[i]; asm volatile ( "fldmias %0, {s0-s11} \n\t" "fldmias %1, {s12-s17} \n\t" "fldmias %2, {s18-s23} \n\t" "fmacs s0, s6, s12 \n\t" "fmuls s6, s6, s18 \n\t" "fstmias %0, {s0-s11} \n\t" : "=r" (p) : "r" (p), "r" (dtArray), "r" (dragArray) : ); } Was: 1.4 seconds
VFP asm for (int i=0; i<iterations; ++i) { Particle2* p = &s_particles2[i]; asm volatile ( "fldmias %0, {s0-s11} \n\t" "fldmias %1, {s12-s17} \n\t" "fldmias %2, {s18-s23} \n\t" "fmacs s0, s6, s12 \n\t" "fmuls s6, s6, s18 \n\t" "fstmias %0, {s0-s11} \n\t" : "=r" (p) : "r" (p), "r" (dtArray), "r" (dragArray) : ); } Was: 1.4 seconds
Now: 1.2 seconds
VFP asmWhat’s the loop/cache overhead?
for (int i=0; i<MaxParticles; ++i) { Particle* p = &s_particles[i]; p->x = p->vx; p->y = p->vy; p->z = p->vz; }
VFP asmWhat’s the loop/cache overhead?
for (int i=0; i<MaxParticles; ++i) { Particle* p = &s_particles[i]; p->x = p->vx; p->y = p->vy; p->z = p->vz; }
Was: 1.2 seconds
VFP asmWhat’s the loop/cache overhead?
for (int i=0; i<MaxParticles; ++i) { Particle* p = &s_particles[i]; p->x = p->vx; p->y = p->vy; p->z = p->vz; }
Was: 1.2 secondsNow: 1.2 seconds!!!!
Matrix multiply
Matrix multiplyStraight from vfpmathlib
Matrix multiply
Touch: 0.037919 s
Straight from vfpmathlib
Matrix multiply
Touch: 0.037919 sNormal: 0.096855 s
Straight from vfpmathlib
Matrix multiply
Touch: 0.037919 sNormal: 0.096855 sVFP: 0.042216 s
Straight from vfpmathlib
Matrix multiply
Touch: 0.037919 sNormal: 0.096855 sVFP: 0.042216 s
About 2x faster!
Straight from vfpmathlib
Good use of vfp
Good use of vfp
• Matrix operations
Good use of vfp
• Matrix operations
• Particle systems
Good use of vfp
• Matrix operations
• Particle systems
• Skinning
Good use of vfp
• Matrix operations
• Particle systems
• Skinning
• Physics
Good use of vfp
• Matrix operations
• Particle systems
• Skinning
• Physics
• Procedural content generation
Good use of vfp
• Matrix operations
• Particle systems
• Skinning
• Physics
• Procedural content generation
• ....
What about the 3GS?
What about the 3GS?
3G 3GS
Thumb
Normal
VFP1
VFP2
Touch
7.2 8.0
2.6 2.6
1.4 1.30
1.2 0.64
1.2 0.18
What about the 3GS?
3G 3GS
Thumb
Normal
VFP1
VFP2
Touch
7.2 8.0
2.6 2.6
1.4 1.30
1.2 0.64
1.2 0.18
What about the 3GS?
3G 3GS
Thumb
Normal
VFP1
VFP2
Touch
7.2 8.0
2.6 2.6
1.4 1.30
1.2 0.64
1.2 0.18
What about the 3GS?
3G 3GS
Thumb
Normal
VFP1
VFP2
Touch
7.2 8.0
2.6 2.6
1.4 1.30
1.2 0.64
1.2 0.18
What about the 3GS?
3G 3GS
Thumb
Normal
VFP1
VFP2
Touch
7.2 8.0
2.6 2.6
1.4 1.30
1.2 0.64
1.2 0.18
What about the 3GS?
3G 3GS
Thumb
Normal
VFP1
VFP2
Touch
7.2 8.0
2.6 2.6
1.4 1.30
1.2 0.64
1.2 0.18
What about the 3GS?
3G 3GS
Thumb
Normal
VFP1
VFP2
Touch
7.2 8.0
2.6 2.6
1.4 1.30
1.2 0.64
1.2 0.18
More 3GS: NEON
More 3GS: NEON
• SIMD coprocessor
More 3GS: NEON
• SIMD coprocessor
• Floating point and integer
More 3GS: NEON
• SIMD coprocessor
• Floating point and integer
• Huge potential
More 3GS: NEON
• SIMD coprocessor
• Floating point and integer
• Huge potential
• Very little documentation right now :-(
Thank you!
Noel LlopisSnappy Touch
http://twitter.com/snappytouchnoel@snappytouch.com
http://gamesfromwithin.com
Recommended