JIT: Investigate fallthrough flow regression from empty block removal

In the diffs from #129092 there are a few cases where we no longer have fallthrough flow.

Example 1. coreclr_tests System.Math:Max.
G_M48648_IG08 -> G_M48648_IG07

G_M48648_IG02:  ;; offset=0x0000
       vucomiss xmm0, xmm1
       jp       SHORT G_M48648_IG03
       je       SHORT G_M48648_IG08
						;; size=8 bbWeight=1 PerfScore 4.00
G_M48648_IG03:  ;; offset=0x0008
       vucomiss xmm0, xmm0
       jp       SHORT G_M48648_IG04
       vucomiss xmm0, xmm1
       jbe      SHORT G_M48648_IG07
						;; size=12 bbWeight=1 PerfScore 6.00
G_M48648_IG04:  ;; offset=0x0014
       vmovaps  xmm1, xmm0
						;; size=4 bbWeight=1 PerfScore 0.25
G_M48648_IG05:  ;; offset=0x0018
       vmovaps  xmm0, xmm1
						;; size=4 bbWeight=1 PerfScore 0.25
G_M48648_IG06:  ;; offset=0x001C
       ret      
						;; size=1 bbWeight=1 PerfScore 1.00
G_M48648_IG07:  ;; offset=0x001D
       jmp      SHORT G_M48648_IG05
						;; size=2 bbWeight=0 PerfScore 0.00
G_M48648_IG08:  ;; offset=0x001F
       vmovd    eax, xmm1
       test     eax, eax
       jl       SHORT G_M48648_IG04
       jmp      SHORT G_M48648_IG07
						;; size=10 bbWeight=0 PerfScore 0.00

Example 2. benchmarks.run_pgo.windows System.Text.RegularExpressions.RegexNode:get_IsBacktrackingConstruct
G_M48811_IG07 -> G_M48811_IG05

G_M48811_IG02:  ;; offset=0x0000
       movzx    rax, byte  ptr [rcx+0x2E]
       lea      edx, [rax-0x03]
       cmp      edx, 5
       jbe      SHORT G_M48811_IG07
       cmp      eax, 24
       je       SHORT G_M48811_IG05
       add      eax, -26
       cmp      eax, 1
       jbe      SHORT G_M48811_IG06
						;; size=25 bbWeight=1 PerfScore 6.50
G_M48811_IG03:  ;; offset=0x0019
       xor      eax, eax
						;; size=2 bbWeight=1 PerfScore 0.25
G_M48811_IG04:  ;; offset=0x001B
       ret      
						;; size=1 bbWeight=1 PerfScore 1.00
G_M48811_IG05:  ;; offset=0x001C
       mov      eax, 1
       jmp      SHORT G_M48811_IG04
						;; size=7 bbWeight=0 PerfScore 0.00
G_M48811_IG06:  ;; offset=0x0023
       mov      eax, dword ptr [rcx+0x20]
       cmp      eax, dword ptr [rcx+0x24]
       je       SHORT G_M48811_IG03
       jmp      SHORT G_M48811_IG05
						;; size=10 bbWeight=0 PerfScore 0.00
G_M48811_IG07:  ;; offset=0x002D
       mov      eax, dword ptr [rcx+0x20]
       cmp      eax, dword ptr [rcx+0x24]
       je       SHORT G_M48811_IG03
       jmp      SHORT G_M48811_IG05
						;; size=10 bbWeight=0 PerfScore 0.00

Note how these cases are evolving arround bbWeight=0.0.
The original PR just removed empty blocks. I don't see why the new codegen would be any worse.

@JulieLeeMSFT

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

I'm a bot. Here is a possible related and/or duplicate issue (I may be wrong):

JIT: fgRemoveBlock does not maintain edge weights #96614

@AndyAyersMS

So the reason for the diffs is quite simple.
Layout opts intentionally don't waste time on bbWeight=0.0 blocks:

runtime/src/coreclr/jit/fgopt.cpp

Lines 4217 to 4228 in 2f38215

    
           // When walking the RPO-based layout, compact the hot blocks, and remember the end of the hot section. 
        
           // We don't want to waste time running 3-opt on cold blocks, or on handler sections. 
        
           unsigned numHotBlocks  = 0; 
        
           auto     addToSequence = [this, initialLayout, &numHotBlocks](BasicBlock* block) { 
        
               // The first block really shouldn't be cold, but if it is, ensure it's still placed first. 
        
               if (!block->hasHndIndex() && (!block->isBBWeightCold(this) || block->IsFirst())) 
        
               { 
        
                   // Set the block's ordinal. 
        
                   block->bbPreorderNum          = numHotBlocks; 
        
                   initialLayout[numHotBlocks++] = block; 
        
               } 
        
           };

I wonder if we can at least do something to avoid spurious diffs. Such as ordering all bbWeight=0.0 blocks by bbNum.
Also I'm gonna try lift that restriction just to see diffs/tp impact. (@AndyAyersMS)

Navigation Menu

Repository navigation

JIT: Investigate fallthrough flow regression from empty block removal #129275

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	// When walking the RPO-based layout, compact the hot blocks, and remember the end of the hot section.
	// We don't want to waste time running 3-opt on cold blocks, or on handler sections.
	unsigned numHotBlocks = 0;
	auto addToSequence = [this, initialLayout, &numHotBlocks](BasicBlock* block) {
	// The first block really shouldn't be cold, but if it is, ensure it's still placed first.
	if (!block->hasHndIndex() && (!block->isBBWeightCold(this) \|\| block->IsFirst()))
	{
	// Set the block's ordinal.
	block->bbPreorderNum = numHotBlocks;
	initialLayout[numHotBlocks++] = block;
	}
	};

JIT: Investigate fallthrough flow regression from empty block removal #129275

Description

Activity

dotnet-policy-service commented on Jun 11, 2026

MihuBot commented on Jun 11, 2026

BoyBaykiller commented on Jun 12, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions