os/kernelhwsrv/kernel/eka/include/x86hlp_gcc.inl
changeset 0 bde4ae8d615e
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/os/kernelhwsrv/kernel/eka/include/x86hlp_gcc.inl	Fri Jun 15 03:10:57 2012 +0200
     1.3 @@ -0,0 +1,455 @@
     1.4 +// Copyright (c) 2007-2009 Nokia Corporation and/or its subsidiary(-ies).
     1.5 +// All rights reserved.
     1.6 +// This component and the accompanying materials are made available
     1.7 +// under the terms of the License "Eclipse Public License v1.0"
     1.8 +// which accompanies this distribution, and is available
     1.9 +// at the URL "http://www.eclipse.org/legal/epl-v10.html".
    1.10 +//
    1.11 +// Initial Contributors:
    1.12 +// Nokia Corporation - initial contribution.
    1.13 +//
    1.14 +// Contributors:
    1.15 +//
    1.16 +// Description:
    1.17 +// e32\common\x86\x86hlp_gcc.inl
    1.18 +// If there are no exports then GCC 3.4.x does not generate a .reloc 
    1.19 +// section, without which rombuild can't relocate the .code section
    1.20 +// to its ROM address. Your ROM then goes boom early in the boot sequence.
    1.21 +// This unused export forces the PE to be generated with a .reloc section.
    1.22 +// 
    1.23 +//
    1.24 +
    1.25 +EXPORT_C void __ignore_this_export()
    1.26 +	{
    1.27 +	}
    1.28 +
    1.29 +static void DivisionByZero()
    1.30 +	{
    1.31 +	asm("int 0");
    1.32 +	}
    1.33 +
    1.34 +extern "C" {
    1.35 +
    1.36 +void __NAKED__ _alloca()
    1.37 +{
    1.38 +	// GCC passes the param in eax and expects no return value
    1.39 +	asm("pop ecx");
    1.40 +	asm("sub esp, eax");
    1.41 +	asm("push ecx");
    1.42 +	asm("ret");
    1.43 +}
    1.44 +
    1.45 +void __NAKED__ _allmul()
    1.46 +//
    1.47 +// Multiply two 64 bit integers returning a 64 bit result
    1.48 +// On entry:
    1.49 +//		[esp+4], [esp+8] = arg 1
    1.50 +//		[esp+12], [esp+16] = arg 1
    1.51 +// Return result in edx:eax
    1.52 +// Remove arguments from stack
    1.53 +//
    1.54 +	{
    1.55 +	asm("mov eax, [esp+4]");		// eax = low1
    1.56 +	asm("mul dword ptr [esp+16]");	// edx:eax = low1*high2
    1.57 +	asm("mov ecx, eax");			// keep low 32 bits of product
    1.58 +	asm("mov eax, [esp+8]");		// eax = high1
    1.59 +	asm("mul dword ptr [esp+12]");	// edx:eax = high1*low2
    1.60 +	asm("add ecx, eax");			// accumulate low 32 bits of product
    1.61 +	asm("mov eax, [esp+4]");		// eax = low1
    1.62 +	asm("mul dword ptr [esp+12]");	// edx:eax = low1*low2
    1.63 +	asm("add edx, ecx");			// add cross terms to high 32 bits
    1.64 +	asm("ret");
    1.65 +	}
    1.66 +
    1.67 +void __NAKED__ udiv64_divby0()
    1.68 +	{
    1.69 +	asm("int 0");					// division by zero exception
    1.70 +	asm("ret");
    1.71 +	}
    1.72 +
    1.73 +__NAKED__ /*LOCAL_C*/ void UDiv64()
    1.74 +	{
    1.75 +	// unsigned divide edx:eax by edi:esi
    1.76 +	// quotient in ebx:eax, remainder in edi:edx
    1.77 +	// ecx, ebp, esi also modified
    1.78 +	asm("test edi, edi");
    1.79 +	asm("jnz short UDiv64a");			// branch if divisor >= 2^32
    1.80 +	asm("test esi, esi");
    1.81 +	asm("jz %a0": : "i"(&DivisionByZero)); // if divisor=0, branch to error routine
    1.82 +	asm("mov ebx, eax");				// ebx=dividend low
    1.83 +	asm("mov eax, edx");				// eax=dividend high
    1.84 +	asm("xor edx, edx");				// edx=0
    1.85 +	asm("div esi");						// quotient high now in eax
    1.86 +	asm("xchg eax, ebx");				// quotient high in ebx, dividend low in eax
    1.87 +	asm("div esi");						// quotient now in ebx:eax, remainder in edi:edx
    1.88 +	asm("ret");
    1.89 +	asm("UDiv64e:");
    1.90 +	asm("xor eax, eax");				// set result to 0xFFFFFFFF
    1.91 +	asm("dec eax");
    1.92 +	asm("jmp short UDiv64f");
    1.93 +	asm("UDiv64a:");
    1.94 +	asm("js short UDiv64b");			// skip if divisor msb set
    1.95 +	asm("bsr ecx, edi");				// ecx=bit number of divisor msb - 32
    1.96 +	asm("inc cl");
    1.97 +	asm("push edi");					// save divisor high
    1.98 +	asm("push esi");					// save divisor low
    1.99 +	asm("shrd esi, edi, cl");			// shift divisor right so that msb is bit 31
   1.100 +	asm("mov ebx, edx");				// dividend into ebx:ebp
   1.101 +	asm("mov ebp, eax");
   1.102 +	asm("shrd eax, edx, cl");			// shift dividend right same number of bits
   1.103 +	asm("shr edx, cl");
   1.104 +	asm("cmp edx, esi");				// check if approx quotient will be 2^32
   1.105 +	asm("jae short UDiv64e");			// if so, true result must be 0xFFFFFFFF
   1.106 +	asm("div esi");						// approximate quotient now in eax
   1.107 +	asm("UDiv64f:");
   1.108 +	asm("mov ecx, eax");				// into ecx
   1.109 +	asm("mul edi");						// multiply approx. quotient by divisor high
   1.110 +	asm("mov esi, eax");				// ls dword into esi, ms into edi
   1.111 +	asm("mov edi, edx");
   1.112 +	asm("mov eax, ecx");				// approx. quotient into eax
   1.113 +	asm("mul dword ptr [esp]");			// multiply approx. quotient by divisor low
   1.114 +	asm("add edx, esi");				// edi:edx:eax now equals approx. quotient * divisor
   1.115 +	asm("adc edi, 0");
   1.116 +	asm("xor esi, esi");
   1.117 +	asm("sub ebp, eax");				// subtract dividend - approx. quotient *divisor
   1.118 +	asm("sbb ebx, edx");
   1.119 +	asm("sbb esi, edi");
   1.120 +	asm("jnc short UDiv64c");			// if no borrow, result OK
   1.121 +	asm("dec ecx");						// else result is one too big
   1.122 +	asm("add ebp, [esp]");				// and add divisor to get correct remainder
   1.123 +	asm("adc ebx, [esp+4]");
   1.124 +	asm("UDiv64c:");
   1.125 +	asm("mov eax, ecx");				// result into ebx:eax, remainder into edi:edx
   1.126 +	asm("mov edi, ebx");
   1.127 +	asm("mov edx, ebp");
   1.128 +	asm("xor ebx, ebx");
   1.129 +	asm("add esp, 8");					// remove temporary values from stack
   1.130 +	asm("ret");
   1.131 +	asm("UDiv64b:");
   1.132 +	asm("mov ebx, 1");
   1.133 +	asm("sub eax, esi");				// subtract divisor from dividend
   1.134 +	asm("sbb edx, edi");
   1.135 +	asm("jnc short UDiv64d");			// if no borrow, result=1, remainder in edx:eax
   1.136 +	asm("add eax, esi");				// else add back
   1.137 +	asm("adc edx, edi");
   1.138 +	asm("dec ebx");						// and decrement quotient
   1.139 +	asm("UDiv64d:");
   1.140 +	asm("mov edi, edx");				// remainder into edi:edx
   1.141 +	asm("mov edx, eax");
   1.142 +	asm("mov eax, ebx");				// result in ebx:eax
   1.143 +	asm("xor ebx, ebx");
   1.144 +	asm("ret");
   1.145 +	}
   1.146 +
   1.147 +__NAKED__ void _aulldvrm()
   1.148 +//
   1.149 +// Divide two 64 bit unsigned integers, returning a 64 bit result
   1.150 +// and a 64 bit remainder
   1.151 +//
   1.152 +// On entry:
   1.153 +//		[esp+4], [esp+8] = dividend
   1.154 +//		[esp+12], [esp+16] = divisor
   1.155 +//
   1.156 +// Return (dividend / divisor) in edx:eax
   1.157 +// Return (dividend % divisor) in ebx:ecx
   1.158 +//
   1.159 +// Remove arguments from stack
   1.160 +//
   1.161 +	{
   1.162 +	asm("push ebp");
   1.163 +	asm("push edi");
   1.164 +	asm("push esi");
   1.165 +	asm("mov eax, [esp+16]");
   1.166 +	asm("mov edx, [esp+20]");
   1.167 +	asm("mov esi, [esp+24]");
   1.168 +	asm("mov edi, [esp+28]");
   1.169 +	asm("call %a0": : "i"(&UDiv64));
   1.170 +	asm("mov ecx, edx");
   1.171 +	asm("mov edx, ebx");
   1.172 +	asm("mov ebx, edi");
   1.173 +	asm("pop esi");
   1.174 +	asm("pop edi");
   1.175 +	asm("pop ebp");
   1.176 +	asm("ret");
   1.177 +	}
   1.178 +
   1.179 +__NAKED__ void _alldvrm()
   1.180 +//
   1.181 +// Divide two 64 bit signed integers, returning a 64 bit result
   1.182 +// and a 64 bit remainder
   1.183 +//
   1.184 +// On entry:
   1.185 +//		[esp+4], [esp+8] = dividend
   1.186 +//		[esp+12], [esp+16] = divisor
   1.187 +//
   1.188 +// Return (dividend / divisor) in edx:eax
   1.189 +// Return (dividend % divisor) in ebx:ecx
   1.190 +//
   1.191 +// Remove arguments from stack
   1.192 +//
   1.193 +	{
   1.194 +	asm("push ebp");
   1.195 +	asm("push edi");
   1.196 +	asm("push esi");
   1.197 +	asm("mov eax, [esp+16]");
   1.198 +	asm("mov edx, [esp+20]");
   1.199 +	asm("mov esi, [esp+24]");
   1.200 +	asm("mov edi, [esp+28]");
   1.201 +	asm("test edx, edx");
   1.202 +	asm("jns alldrvm_dividend_nonnegative");
   1.203 +	asm("neg edx");
   1.204 +	asm("neg eax");
   1.205 +	asm("sbb edx, 0");
   1.206 +	asm("alldrvm_dividend_nonnegative:");
   1.207 +	asm("test edi, edi");
   1.208 +	asm("jns alldrvm_divisor_nonnegative");
   1.209 +	asm("neg edi");
   1.210 +	asm("neg esi");
   1.211 +	asm("sbb edi, 0");
   1.212 +	asm("alldrvm_divisor_nonnegative:");
   1.213 +	asm("call %a0": : "i"(&UDiv64));
   1.214 +	asm("mov ebp, [esp+20]");
   1.215 +	asm("mov ecx, edx");
   1.216 +	asm("xor ebp, [esp+28]");
   1.217 +	asm("mov edx, ebx");
   1.218 +	asm("mov ebx, edi");
   1.219 +	asm("jns alldrvm_quotient_nonnegative");
   1.220 +	asm("neg edx");
   1.221 +	asm("neg eax");
   1.222 +	asm("sbb edx, 0");
   1.223 +	asm("alldrvm_quotient_nonnegative:");
   1.224 +	asm("cmp dword ptr [esp+20], 0");
   1.225 +	asm("jns alldrvm_rem_nonnegative");
   1.226 +	asm("neg ebx");
   1.227 +	asm("neg ecx");
   1.228 +	asm("sbb ebx, 0");
   1.229 +	asm("alldrvm_rem_nonnegative:");
   1.230 +	asm("pop esi");
   1.231 +	asm("pop edi");
   1.232 +	asm("pop ebp");
   1.233 +	asm("ret");
   1.234 +	}
   1.235 +
   1.236 +//__NAKED__ void _aulldiv()
   1.237 +__NAKED__ void __udivdi3 ()
   1.238 +//
   1.239 +// Divide two 64 bit unsigned integers returning a 64 bit result
   1.240 +// On entry:
   1.241 +//		[esp+4], [esp+8] = dividend
   1.242 +//		[esp+12], [esp+16] = divisor
   1.243 +// Return result in edx:eax
   1.244 +// Remove arguments from stack
   1.245 +//
   1.246 +	{
   1.247 +	asm("push ebp");
   1.248 +	asm("push edi");
   1.249 +	asm("push esi");
   1.250 +	asm("push ebx");
   1.251 +	asm("mov eax, [esp+20]");
   1.252 +	asm("mov edx, [esp+24]");
   1.253 +	asm("mov esi, [esp+28]");
   1.254 +	asm("mov edi, [esp+32]");
   1.255 +	asm("call %a0": : "i"(&UDiv64));
   1.256 +	asm("mov edx, ebx");
   1.257 +	asm("pop ebx");
   1.258 +	asm("pop esi");
   1.259 +	asm("pop edi");
   1.260 +	asm("pop ebp");
   1.261 +	asm("ret");
   1.262 +	}
   1.263 +
   1.264 +
   1.265 +__NAKED__ void __divdi3()
   1.266 +
   1.267 +//
   1.268 +// Divide two 64 bit signed integers returning a 64 bit result
   1.269 +// On entry:
   1.270 +//		[esp+4], [esp+8] = dividend
   1.271 +//		[esp+12], [esp+16] = divisor
   1.272 +// Return result in edx:eax
   1.273 +// Remove arguments from stack
   1.274 +//
   1.275 +	{
   1.276 +	asm("push ebp");
   1.277 +	asm("push edi");
   1.278 +	asm("push esi");
   1.279 +	asm("push ebx");
   1.280 +	asm("mov eax, [esp+20]");
   1.281 +	asm("mov edx, [esp+24]");
   1.282 +	asm("mov esi, [esp+28]");
   1.283 +	asm("mov edi, [esp+32]");
   1.284 +	asm("test edx, edx");
   1.285 +	asm("jns divdi_dividend_nonnegative");
   1.286 +	asm("neg edx");
   1.287 +	asm("neg eax");
   1.288 +	asm("sbb edx, 0");
   1.289 +	asm("divdi_dividend_nonnegative:");
   1.290 +	asm("test edi, edi");
   1.291 +	asm("jns divdi_divisor_nonnegative");
   1.292 +	asm("neg edi");
   1.293 +	asm("neg esi");
   1.294 +	asm("sbb edi, 0");
   1.295 +	asm("divdi_divisor_nonnegative:");
   1.296 +	asm("call %a0": : "i"(&UDiv64));
   1.297 +	asm("mov ecx, [esp+24]");
   1.298 +	asm("mov edx, ebx");
   1.299 +	asm("xor ecx, [esp+32]");
   1.300 +	asm("jns divdi_quotient_nonnegative");
   1.301 +	asm("neg edx");
   1.302 +	asm("neg eax");
   1.303 +	asm("sbb edx, 0");
   1.304 +	asm("divdi_quotient_nonnegative:");
   1.305 +	asm("pop ebx");
   1.306 +	asm("pop esi");
   1.307 +	asm("pop edi");
   1.308 +	asm("pop ebp");
   1.309 +	asm("ret");
   1.310 +	}
   1.311 +
   1.312 +__NAKED__ void __umoddi3()
   1.313 +//
   1.314 +// Divide two 64 bit unsigned integers and return 64 bit remainder
   1.315 +// On entry:
   1.316 +//		[esp+4], [esp+8] = dividend
   1.317 +//		[esp+12], [esp+16] = divisor
   1.318 +// Return result in edx:eax
   1.319 +// Remove arguments from stack
   1.320 +//
   1.321 +	{
   1.322 +	asm("push ebp");
   1.323 +	asm("push edi");
   1.324 +	asm("push esi");
   1.325 +	asm("push ebx");
   1.326 +	asm("mov eax, [esp+20]");
   1.327 +	asm("mov edx, [esp+24]");
   1.328 +	asm("mov esi, [esp+28]");
   1.329 +	asm("mov edi, [esp+32]");
   1.330 +	asm("call %a0": : "i"(&UDiv64));
   1.331 +	asm("mov eax, edx");
   1.332 +	asm("mov edx, edi");
   1.333 +	asm("pop ebx");
   1.334 +	asm("pop esi");
   1.335 +	asm("pop edi");
   1.336 +	asm("pop ebp");
   1.337 +	asm("ret");
   1.338 +	}
   1.339 +
   1.340 +__NAKED__ void __moddi3()
   1.341 +//
   1.342 +// Divide two 64 bit signed integers and return 64 bit remainder
   1.343 +// On entry:
   1.344 +//		[esp+4], [esp+8] = dividend
   1.345 +//		[esp+12], [esp+16] = divisor
   1.346 +// Return result in edx:eax
   1.347 +// Remove arguments from stack
   1.348 +//
   1.349 +	{
   1.350 +	asm("push ebp");
   1.351 +	asm("push edi");
   1.352 +	asm("push esi");
   1.353 +	asm("push ebx");
   1.354 +	asm("mov eax, [esp+20]");
   1.355 +	asm("mov edx, [esp+24]");
   1.356 +	asm("mov esi, [esp+28]");
   1.357 +	asm("mov edi, [esp+32]");
   1.358 +	asm("test edx, edx");
   1.359 +	asm("jns dividend_nonnegative");
   1.360 +	asm("neg edx");
   1.361 +	asm("neg eax");
   1.362 +	asm("sbb edx, 0");
   1.363 +	asm("dividend_nonnegative:");
   1.364 +	asm("test edi, edi");
   1.365 +	asm("jns divisor_nonnegative");
   1.366 +	asm("neg edi");
   1.367 +	asm("neg esi");
   1.368 +	asm("sbb edi, 0");
   1.369 +	asm("divisor_nonnegative:");
   1.370 +	asm("call %a0": : "i"(&UDiv64));
   1.371 +	asm("mov eax, edx");
   1.372 +	asm("mov edx, edi");
   1.373 +	asm("cmp dword ptr [esp+24], 0");
   1.374 +	asm("jns rem_nonnegative");
   1.375 +	asm("neg edx");
   1.376 +	asm("neg eax");
   1.377 +	asm("sbb edx, 0");
   1.378 +	asm("rem_nonnegative:");
   1.379 +	asm("pop ebx");
   1.380 +	asm("pop esi");
   1.381 +	asm("pop edi");
   1.382 +	asm("pop ebp");
   1.383 +	asm("ret");
   1.384 +	}
   1.385 +
   1.386 +__NAKED__ void _allshr()
   1.387 +//
   1.388 +// Arithmetic shift right EDX:EAX by CL
   1.389 +//
   1.390 +	{
   1.391 +	asm("cmp cl, 64");
   1.392 +	asm("jae asr_count_ge_64");
   1.393 +	asm("cmp cl, 32");
   1.394 +	asm("jae asr_count_ge_32");
   1.395 +	asm("shrd eax, edx, cl");
   1.396 +	asm("sar edx, cl");
   1.397 +	asm("ret");
   1.398 +	asm("asr_count_ge_32:");
   1.399 +	asm("sub cl, 32");
   1.400 +	asm("mov eax, edx");
   1.401 +	asm("cdq");
   1.402 +	asm("sar eax, cl");
   1.403 +	asm("ret");
   1.404 +	asm("asr_count_ge_64:");
   1.405 +	asm("sar edx, 32");
   1.406 +	asm("mov eax, edx");
   1.407 +	asm("ret");
   1.408 +	}
   1.409 +
   1.410 +__NAKED__ void _allshl()
   1.411 +//
   1.412 +// shift left EDX:EAX by CL
   1.413 +//
   1.414 +	{
   1.415 +	asm("cmp cl, 64");
   1.416 +	asm("jae lsl_count_ge_64");
   1.417 +	asm("cmp cl, 32");
   1.418 +	asm("jae lsl_count_ge_32");
   1.419 +	asm("shld edx, eax, cl");
   1.420 +	asm("shl eax, cl");
   1.421 +	asm("ret");
   1.422 +	asm("lsl_count_ge_32:");
   1.423 +	asm("sub cl, 32");
   1.424 +	asm("mov edx, eax");
   1.425 +	asm("xor eax, eax");
   1.426 +	asm("shl edx, cl");
   1.427 +	asm("ret");
   1.428 +	asm("lsl_count_ge_64:");
   1.429 +	asm("xor edx, edx");
   1.430 +	asm("xor eax, eax");
   1.431 +	asm("ret");
   1.432 +	}
   1.433 +
   1.434 +__NAKED__ void _aullshr()
   1.435 +//
   1.436 +// Logical shift right EDX:EAX by CL
   1.437 +//
   1.438 +	{
   1.439 +	asm("cmp cl, 64");
   1.440 +	asm("jae lsr_count_ge_64");
   1.441 +	asm("cmp cl, 32");
   1.442 +	asm("jae lsr_count_ge_32");
   1.443 +	asm("shrd eax, edx, cl");
   1.444 +	asm("shr edx, cl");
   1.445 +	asm("ret");
   1.446 +	asm("lsr_count_ge_32:");
   1.447 +	asm("sub cl, 32");
   1.448 +	asm("mov eax, edx");
   1.449 +	asm("xor edx, edx");
   1.450 +	asm("shr eax, cl");
   1.451 +	asm("ret");
   1.452 +	asm("lsr_count_ge_64:");
   1.453 +	asm("xor edx, edx");
   1.454 +	asm("xor eax, eax");
   1.455 +	asm("ret");
   1.456 +	}
   1.457 +
   1.458 +}