How does Ibex support F instructions??!!!!!

[Kian75]

Hi... I have been working on Ibex core for a short time and I know that Ibex is 32IMC but I tested a code to multiply two float numbers just to make sure!! Actually I used this git hub page that use Ibex as their core: https://github.com/pbing/ibex_wb I programmed the FPGA board and it works...It actually multiplies 2.5 and 2 as float numbers and the output of multiplication is 5 (I can see the result on led)...Here is the code: // Copyright lowRISC contributors. // Licensed under the Apache License, Version 2.0, see LICENSE for details. // SPDX-License-Identifier: Apache-2.0

include

define CLK_FIXED_FREQ_HZ (25ULL 1000 1000)

/**

• Delay loop executing within 8 cycles on ibex / static void delay_loop_ibex(unsigned long loops) { int out; / only to notify compiler of modifications to |loops| */ asm volatile( "1: nop \n" // 1 cycle " nop \n" // 1 cycle " nop \n" // 1 cycle " nop \n" // 1 cycle " addi %1, %1, -1 \n" // 1 cycle " bnez %1, 1b \n" // 3 cycles : "=&r" (out) : "0" (loops) ); }

static int usleep_ibex(unsigned long usec) { unsigned long usec_cycles; usec_cycles = CLK_FIXED_FREQ_HZ * usec / 1000 / 1000 / 8;

delay_loop_ibex(usec_cycles); return 0; }

static int usleep(unsigned long usec) { return usleep_ibex(usec); }

int main(int argc, char *argv) { volatile uint32_t var = (volatile uint32_t ) 0x10000000; float x=2.5; float y=2; var = x*y; } After compiling I get this from compiler for the main function: As you can see there are some F instructions.... 00000000000101d0

: 101d0: 7179 addi sp,sp,-48 101d2: f422 sd s0,40(sp) 101d4: 1800 addi s0,sp,48 101d6: 87aa mv a5,a0 101d8: fcb43823 sd a1,-48(s0) 101dc: fcf42e23 sw a5,-36(s0) 101e0: 100007b7 lui a5,0x10000 101e4: fef43423 sd a5,-24(s0) 101e8: 67c1 lui a5,0x10 101ea: 4f07a787 flw fa5,1264(a5) # 104f0 <errno+0x6> 101ee: fef42227 fsw fa5,-28(s0) 101f2: 67c1 lui a5,0x10 101f4: 4f47a787 flw fa5,1268(a5) # 104f4 <errno+0xa> 101f8: fef42027 fsw fa5,-32(s0) 101fc: fe442707 flw fa4,-28(s0) 10200: fe042787 flw fa5,-32(s0) 10204: 10f777d3 fmul.s fa5,fa4,fa5 10208: c01797d3 fcvt.wu.s a5,fa5,rtz 1020c: 0007871b sext.w a4,a5 10210: fe843783 ld a5,-24(s0) 10214: c398 sw a4,0(a5) 10216: 4781 li a5,0 10218: 853e mv a0,a5 1021a: 7422 ld s0,40(sp) 1021c: 6145 addi sp,sp,48 1021e: 8082 ret It is a bit awkward but how is it possible? How does Ibex works with this F instructions??

[imphil]

No, Ibex doesn't support the F extension. Your compiler produced an instruction stream with F instructions in there, but Ibex won't be able to execute it, unless you (aka the compiler) provides an implementation for it. That's called "soft float", i.e. an implementation of floating point arithmetic in software.

Try it out by providing the corresponding "-march" flags to GCC, e.g. -march=rv32imf or -march=rv32im.

[Kian75]

Thank you for your answer... I am really confused so sorry if I am asking again... I am trying to understand that soft float part that you mentioned and my mind is blowing :(( ... I used this following configurations: About the compiler I used https://github.com/riscv/riscv-gnu-toolchain and I installed the (Newlib/Linux multilib)... For compiling the code I used this make file https://github.com/pbing/ibex_wb/blob/master/soc/fpga/arty-a7-100/sw/led/Makefile in this make file the -march=rv32im ... So you say that with this configuration for compiler Ibex can implement "soft float?!" with the help of compiler?

[tomeroberts]

So you say that with this configuration for compiler Ibex can implement "soft float?!" with the help of compiler?

That's correct. The compiler knows that the CPU can't execute floating point arithmetic so it emulates that behavior by inserting routines based on integer arithmetic.

As a more simple example, imagine a basic CPU that doesn't have a multiply instruction. You could create a function to multiply two numbers using shift and addition instructions, then substitute that in anywhere you want to multiply. What the compiler does for soft-float is (sort-of) like that.

[Kian75]

Now I get it... Thank you so much for your time and your easy to understand answer... :))