[RISC-V] Reverse unordered FP comparisons in branches #115943
+37
−56
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
dotnet-policy-service
bot
added
the
community-contribution
github-actions
bot
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
|
No regressions. Diffs are based on 94,874 contexts (52,265 MinOpts, 42,609 FullOpts). MISSED contexts: 3 (0.00%) Overall (-23,124 bytes)
MinOpts (-604 bytes)
FullOpts (-22,520 bytes)
Example diffslinux.riscv64.Checked.mch-12 (-11.11%) : 86422.dasm - Microsoft.VisualBasic.CompilerServices.Operators:CompareSingle(float,float):int (FullOpts)@@ -23,22 +23,19 @@ G_M44467_IG01: ; bbWeight=1, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref,
;; size=16 bbWeight=1 PerfScore 9.00
G_M44467_IG02: ; bbWeight=1, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
feq.s a0, fa0, fa1
- xori a0, a0, 1
sext.w t6, a0
- beqz t6, G_M44467_IG06
- ;; size=16 bbWeight=1 PerfScore 7.50
+ bnez t6, G_M44467_IG06
+ ;; size=12 bbWeight=1 PerfScore 7.00
G_M44467_IG03: ; bbWeight=0.50, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
flt.s a0, fa0, fa1
- xori a0, a0, 1
sext.w t6, a0
- beqz t6, G_M44467_IG05
+ bnez t6, G_M44467_IG05
flt.s a0, fa1, fa0
- xori a0, a0, 1
sext.w t6, a0
- beqz t6, G_M44467_IG04
+ bnez t6, G_M44467_IG04
addi a0, zero, 0xD1FFAB1E
j G_M44467_IG07
- ;; size=40 bbWeight=0.50 PerfScore 8.50
+ ;; size=32 bbWeight=0.50 PerfScore 8.00
G_M44467_IG04: ; bbWeight=0.50, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
addi a0, zero, 0xD1FFAB1E
j G_M44467_IG07
@@ -56,7 +53,7 @@ G_M44467_IG07: ; bbWeight=1, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref,
addi sp, sp, 16
ret ;; size=16 bbWeight=1 PerfScore 7.50
-; Total bytes of code 108, prolog size 16, PerfScore 34.75, instruction count 27, allocated bytes for code 108 (MethodHash=6af2524c) for method Microsoft.VisualBasic.CompilerServices.Operators:CompareSingle(float,float):int (FullOpts)
+; Total bytes of code 96, prolog size 16, PerfScore 33.75, instruction count 24, allocated bytes for code 96 (MethodHash=6af2524c) for method Microsoft.VisualBasic.CompilerServices.Operators:CompareSingle(float,float):int (FullOpts)
; ============================================================
Unwind Info:
@@ -67,7 +64,7 @@ Unwind Info:
E bit : 0
X bit : 0
Vers : 0
- Function Length : 27 (0x0001b) Actual length = 108 (0x00006c)
+ Function Length : 24 (0x00018) Actual length = 96 (0x000060)
---- Epilog scopes ----
---- Scope 0
Epilog Start Offset : 3523193630 (0xd1ffab1e) Actual offset = 3523193630 (0xd1ffab1e) Offset from main function begin = 3523193630 (0xd1ffab1e)-12 (-11.11%) : 86423.dasm - Microsoft.VisualBasic.CompilerServices.Operators:CompareDouble(double,double):int (FullOpts)@@ -23,22 +23,19 @@ G_M37116_IG01: ; bbWeight=1, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref,
;; size=16 bbWeight=1 PerfScore 9.00
G_M37116_IG02: ; bbWeight=1, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
feq.d a0, fa0, fa1
- xori a0, a0, 1
sext.w t6, a0
- beqz t6, G_M37116_IG06
- ;; size=16 bbWeight=1 PerfScore 7.50
+ bnez t6, G_M37116_IG06
+ ;; size=12 bbWeight=1 PerfScore 7.00
G_M37116_IG03: ; bbWeight=0.50, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
flt.d a0, fa0, fa1
- xori a0, a0, 1
sext.w t6, a0
- beqz t6, G_M37116_IG05
+ bnez t6, G_M37116_IG05
flt.d a0, fa1, fa0
- xori a0, a0, 1
sext.w t6, a0
- beqz t6, G_M37116_IG04
+ bnez t6, G_M37116_IG04
addi a0, zero, 0xD1FFAB1E
j G_M37116_IG07
- ;; size=40 bbWeight=0.50 PerfScore 8.50
+ ;; size=32 bbWeight=0.50 PerfScore 8.00
G_M37116_IG04: ; bbWeight=0.50, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
addi a0, zero, 0xD1FFAB1E
j G_M37116_IG07
@@ -56,7 +53,7 @@ G_M37116_IG07: ; bbWeight=1, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref,
addi sp, sp, 16
ret ;; size=16 bbWeight=1 PerfScore 7.50
-; Total bytes of code 108, prolog size 16, PerfScore 34.75, instruction count 27, allocated bytes for code 108 (MethodHash=2f356f03) for method Microsoft.VisualBasic.CompilerServices.Operators:CompareDouble(double,double):int (FullOpts)
+; Total bytes of code 96, prolog size 16, PerfScore 33.75, instruction count 24, allocated bytes for code 96 (MethodHash=2f356f03) for method Microsoft.VisualBasic.CompilerServices.Operators:CompareDouble(double,double):int (FullOpts)
; ============================================================
Unwind Info:
@@ -67,7 +64,7 @@ Unwind Info:
E bit : 0
X bit : 0
Vers : 0
- Function Length : 27 (0x0001b) Actual length = 108 (0x00006c)
+ Function Length : 24 (0x00018) Actual length = 96 (0x000060)
---- Epilog scopes ----
---- Scope 0
Epilog Start Offset : 3523193630 (0xd1ffab1e) Actual offset = 3523193630 (0xd1ffab1e) Offset from main function begin = 3523193630 (0xd1ffab1e)-16 (-10.26%) : 47378.dasm - BringUpTest_JTrueGtFP:JTrueGtFP(float):int (FullOpts)@@ -25,31 +25,27 @@ G_M56740_IG02: ; bbWeight=1, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
lui a1, 0xD1FFAB1E
fmv.w.x ft4, a1
flt.s a1, ft4, fa0
- xori a1, a1, 1
sext.w t6, a1
- beqz t6, G_M56740_IG06
- ;; size=28 bbWeight=1 PerfScore 9.50
+ bnez t6, G_M56740_IG06
+ ;; size=24 bbWeight=1 PerfScore 9.00
G_M56740_IG03: ; bbWeight=0.50, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
fmv.w.x ft4, zero
flt.s a1, ft4, fa0
- xori a1, a1, 1
sext.w t6, a1
- beqz t6, G_M56740_IG05
+ bnez t6, G_M56740_IG05
lui a1, 0xD1FFAB1E
fmv.w.x ft4, a1
flt.s a1, ft4, fa0
- xori a1, a1, 1
sext.w t6, a1
- beqz t6, G_M56740_IG04
+ bnez t6, G_M56740_IG04
auipc t6, 0xD1FFAB1E
flw ft4, 0xD1FFAB1E(t6)
flt.s a1, ft4, fa0
- xori a1, a1, 1
sext.w t6, a1
- bnez t6, G_M56740_IG07
+ beqz t6, G_M56740_IG07
addi a0, zero, 0xD1FFAB1E
j G_M56740_IG07
- ;; size=76 bbWeight=0.50 PerfScore 14.50
+ ;; size=64 bbWeight=0.50 PerfScore 13.75
G_M56740_IG04: ; bbWeight=0.50, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref
addi a0, zero, 0xD1FFAB1E
j G_M56740_IG07
@@ -69,7 +65,7 @@ G_M56740_IG07: ; bbWeight=1, gcrefRegs=0000 {}, byrefRegs=0000 {}, byref,
RWD00 dd FF7FFFFFh ; -3.40282e+38
-; Total bytes of code 156, prolog size 16, PerfScore 42.75, instruction count 38, allocated bytes for code 156 (MethodHash=4c11225b) for method BringUpTest_JTrueGtFP:JTrueGtFP(float):int (FullOpts)
+; Total bytes of code 140, prolog size 16, PerfScore 41.50, instruction count 34, allocated bytes for code 140 (MethodHash=4c11225b) for method BringUpTest_JTrueGtFP:JTrueGtFP(float):int (FullOpts)
; ============================================================
Unwind Info:
@@ -80,7 +76,7 @@ Unwind Info:
E bit : 0
X bit : 0
Vers : 0
- Function Length : 39 (0x00027) Actual length = 156 (0x00009c)
+ Function Length : 35 (0x00023) Actual length = 140 (0x00008c)
---- Epilog scopes ----
---- Scope 0
Epilog Start Offset : 3523193630 (0xd1ffab1e) Actual offset = 3523193630 (0xd1ffab1e) Offset from main function begin = 3523193630 (0xd1ffab1e)+0 (0.00%) : 94768.dasm - Microsoft.Diagnostics.Tracing.ZippedETLWriter:PrepForWrite():System.Collections.Generic.List`1[System.String]:this (FullOpts)No diffs found? +0 (0.00%) : 94736.dasm - Microsoft.Diagnostics.Tracing.TraceEventNativeMethods:IsElevated():System.Nullable`1[ubyte] (FullOpts)No diffs found? +0 (0.00%) : 94480.dasm - Microsoft.Diagnostics.Tracing.EventPipeEventMetaDataHeader:.ctor(FastSerialization.PinnedStreamReader,int,int,int,int,int,System.String):this (FullOpts)No diffs found? DetailsSize improvements/regressions per collection
PerfScore improvements/regressions per collection
Context information
jit-analyze output |
d609565 is being scheduled for building and testingGIT: |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Unordered floating-point comparisons are achieved by neg'ing the ordered counterparts. Avoid that by reversing both the FP comparison and the zero-comparison fused with the branch.
Also, minor codegen haircut.
Part of #84834, cc @dotnet/samsung