fix the regs conflicts within the addResolution for LoongArch64 and RISC-V.

[shushanhf]

This PR is part of the issue #69705 to amend the LA's port.

The LoongArch64 is very different with the AArch64 and AMD64.

// For AArch64 and AMD64:                       |  // For LoongArch64
// cmp $reg1, $reg2  <--just compare.           |  the compare and branch can be written in one instruction.
// branch.condition  <--the condition is here.  |  so the comparing registers should be alive for branch.

If insert move before the branch just liking AArch64 or AMD64, there maybe conflicts.

This should be also valid for RISCV64. cc @clamp03

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This PR is part of the issue #69705 to amend the LA's port.

The LoongArch64 is very different with the AArch64 and AMD64.

// For AArch64 and AMD64:                       |  // For LoongArch64
// cmp $reg1, $reg2  <--just compare.           |  the compare and branch can be written in one instruction.
// branch.condition  <--the condition is here.  |  so the comparing registers should be alive for branch.

If insert move before the branch just liking AArch64 or AMD64, there maybe conflicts.

Author:	shushanhf
Assignees:	-
Labels:	area-CodeGen-coreclr, community-contribution
Milestone:	-

[shushanhf]

Hi, @jakobbotsch
Could you please review this PR?
Thanks

[jakobbotsch]

@shushanhf This should be fixed in LSRA instead. It is what this TODO is about:

runtime/src/coreclr/jit/lsra.cpp

Lines 8172 to 8176 in 7486bde

	#if !defined(TARGET_LOONGARCH64) && !defined(TARGET_RISCV64)
	// TODO-LOONGARCH64: Take into account that on LA64, the second
	// operand of a JCMP can be in a register too.
	assert(!lastNode->OperIs(GT_JCMP, GT_JTEST) || lastNode->gtGetOp2()->isContained());
	#endif

[shushanhf]

@shushanhf This should be fixed in LSRA instead. It is what this TODO is about:

runtime/src/coreclr/jit/lsra.cpp

Lines 8172 to 8176 in 7486bde

	#if !defined(TARGET_LOONGARCH64) && !defined(TARGET_RISCV64)
	// TODO-LOONGARCH64: Take into account that on LA64, the second
	// operand of a JCMP can be in a register too.
	assert(!lastNode->OperIs(GT_JCMP, GT_JTEST) || lastNode->gtGetOp2()->isContained());
	#endif

OK, Thanks.
I will amend it for LA.

[jakobbotsch]

Test failure is #86233 that is known by build analysis.