[llvm] b24edff - [X86] AMD Zen 3: same-reg PCMPEQ is an MMX all-ones dep breaking idiom
Roman Lebedev via llvm-commits
llvm-commits at lists.llvm.org
Mon May 10 13:49:57 PDT 2021
Author: Roman Lebedev
Date: 2021-05-10T23:49:26+03:00
New Revision: b24edfff4fb16549b3e5ec434ca79dd86fdb4e43
URL: https://github.com/llvm/llvm-project/commit/b24edfff4fb16549b3e5ec434ca79dd86fdb4e43
DIFF: https://github.com/llvm/llvm-project/commit/b24edfff4fb16549b3e5ec434ca79dd86fdb4e43.diff
LOG: [X86] AMD Zen 3: same-reg PCMPEQ is an MMX all-ones dep breaking idiom
They are, however, not zero-cycle, and do actually execute.
As measured by exegesis, and confirmed by ref docs.
Added:
Modified:
llvm/lib/Target/X86/X86ScheduleZnver3.td
llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-mmx.s
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86ScheduleZnver3.td b/llvm/lib/Target/X86/X86ScheduleZnver3.td
index f377fb90d5fd..43a23c8ab88d 100644
--- a/llvm/lib/Target/X86/X86ScheduleZnver3.td
+++ b/llvm/lib/Target/X86/X86ScheduleZnver3.td
@@ -1547,7 +1547,10 @@ def : IsDepBreakingFunction<[
CMP32rr, CMP32rr_REV,
CMP64rr, CMP64rr_REV ], CheckSameRegOperand<0, 1> >,
-
+ // MMX
+ DepBreakingClass<[
+ MMX_PCMPEQBirr, MMX_PCMPEQWirr, MMX_PCMPEQDirr
+ ], ZeroIdiomPredicate>,
]>;
} // SchedModel
diff --git a/llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-mmx.s b/llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-mmx.s
index f339f692efa1..5909af853dd5 100644
--- a/llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-mmx.s
+++ b/llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-mmx.s
@@ -20,12 +20,12 @@ paddd %mm0, %mm0
# CHECK: Iterations: 1000
# CHECK-NEXT: Instructions: 2000
-# CHECK-NEXT: Total Cycles: 2003
+# CHECK-NEXT: Total Cycles: 504
# CHECK-NEXT: Total uOps: 2000
# CHECK: Dispatch Width: 6
-# CHECK-NEXT: uOps Per Cycle: 1.00
-# CHECK-NEXT: IPC: 1.00
+# CHECK-NEXT: uOps Per Cycle: 3.97
+# CHECK-NEXT: IPC: 3.97
# CHECK-NEXT: Block RThroughput: 0.5
# CHECK: Instruction Info:
@@ -42,12 +42,12 @@ paddd %mm0, %mm0
# CHECK: Register File statistics:
# CHECK-NEXT: Total number of mappings created: 2000
-# CHECK-NEXT: Max number of mappings used: 66
+# CHECK-NEXT: Max number of mappings used: 74
# CHECK: * Register File #1 -- Zn3FpPRF:
# CHECK-NEXT: Number of physical registers: 160
# CHECK-NEXT: Total number of mappings created: 2000
-# CHECK-NEXT: Max number of mappings used: 66
+# CHECK-NEXT: Max number of mappings used: 74
# CHECK: * Register File #2 -- Zn3IntegerPRF:
# CHECK-NEXT: Number of physical registers: 192
@@ -85,16 +85,16 @@ paddd %mm0, %mm0
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12.0] [12.1] [13] [14.0] [14.1] [14.2] [15.0] [15.1] [15.2] [16.0] [16.1] Instructions:
-# CHECK-NEXT: - - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - pcmpeqb %mm0, %mm0
-# CHECK-NEXT: - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - - paddb %mm0, %mm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 0.50 - - - - - - - - - - - - pcmpeqb %mm0, %mm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 - 0.50 - - - - - - - - - - - paddb %mm0, %mm0
# CHECK: Timeline view:
-# CHECK-NEXT: Index 0123456
+# CHECK-NEXT: Index 01234
-# CHECK: [0,0] DeER .. pcmpeqb %mm0, %mm0
-# CHECK-NEXT: [0,1] D=eER.. paddb %mm0, %mm0
-# CHECK-NEXT: [1,0] D==eER. pcmpeqb %mm0, %mm0
-# CHECK-NEXT: [1,1] D===eER paddb %mm0, %mm0
+# CHECK: [0,0] DeER. pcmpeqb %mm0, %mm0
+# CHECK-NEXT: [0,1] D=eER paddb %mm0, %mm0
+# CHECK-NEXT: [1,0] DeE-R pcmpeqb %mm0, %mm0
+# CHECK-NEXT: [1,1] D=eER paddb %mm0, %mm0
# CHECK: Average Wait times (based on the timeline view):
# CHECK-NEXT: [0]: Executions
@@ -103,20 +103,20 @@ paddd %mm0, %mm0
# CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
# CHECK: [0] [1] [2] [3]
-# CHECK-NEXT: 0. 2 2.0 0.5 0.0 pcmpeqb %mm0, %mm0
-# CHECK-NEXT: 1. 2 3.0 0.0 0.0 paddb %mm0, %mm0
-# CHECK-NEXT: 2 2.5 0.3 0.0 <total>
+# CHECK-NEXT: 0. 2 1.0 1.0 0.5 pcmpeqb %mm0, %mm0
+# CHECK-NEXT: 1. 2 2.0 0.0 0.0 paddb %mm0, %mm0
+# CHECK-NEXT: 2 1.5 0.5 0.3 <total>
# CHECK: [1] Code Region
# CHECK: Iterations: 1000
# CHECK-NEXT: Instructions: 2000
-# CHECK-NEXT: Total Cycles: 2003
+# CHECK-NEXT: Total Cycles: 504
# CHECK-NEXT: Total uOps: 2000
# CHECK: Dispatch Width: 6
-# CHECK-NEXT: uOps Per Cycle: 1.00
-# CHECK-NEXT: IPC: 1.00
+# CHECK-NEXT: uOps Per Cycle: 3.97
+# CHECK-NEXT: IPC: 3.97
# CHECK-NEXT: Block RThroughput: 0.5
# CHECK: Instruction Info:
@@ -133,12 +133,12 @@ paddd %mm0, %mm0
# CHECK: Register File statistics:
# CHECK-NEXT: Total number of mappings created: 2000
-# CHECK-NEXT: Max number of mappings used: 66
+# CHECK-NEXT: Max number of mappings used: 74
# CHECK: * Register File #1 -- Zn3FpPRF:
# CHECK-NEXT: Number of physical registers: 160
# CHECK-NEXT: Total number of mappings created: 2000
-# CHECK-NEXT: Max number of mappings used: 66
+# CHECK-NEXT: Max number of mappings used: 74
# CHECK: * Register File #2 -- Zn3IntegerPRF:
# CHECK-NEXT: Number of physical registers: 192
@@ -176,16 +176,16 @@ paddd %mm0, %mm0
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12.0] [12.1] [13] [14.0] [14.1] [14.2] [15.0] [15.1] [15.2] [16.0] [16.1] Instructions:
-# CHECK-NEXT: - - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - pcmpeqw %mm0, %mm0
-# CHECK-NEXT: - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - - paddw %mm0, %mm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 0.50 - - - - - - - - - - - - pcmpeqw %mm0, %mm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 - 0.50 - - - - - - - - - - - paddw %mm0, %mm0
# CHECK: Timeline view:
-# CHECK-NEXT: Index 0123456
+# CHECK-NEXT: Index 01234
-# CHECK: [0,0] DeER .. pcmpeqw %mm0, %mm0
-# CHECK-NEXT: [0,1] D=eER.. paddw %mm0, %mm0
-# CHECK-NEXT: [1,0] D==eER. pcmpeqw %mm0, %mm0
-# CHECK-NEXT: [1,1] D===eER paddw %mm0, %mm0
+# CHECK: [0,0] DeER. pcmpeqw %mm0, %mm0
+# CHECK-NEXT: [0,1] D=eER paddw %mm0, %mm0
+# CHECK-NEXT: [1,0] DeE-R pcmpeqw %mm0, %mm0
+# CHECK-NEXT: [1,1] D=eER paddw %mm0, %mm0
# CHECK: Average Wait times (based on the timeline view):
# CHECK-NEXT: [0]: Executions
@@ -194,20 +194,20 @@ paddd %mm0, %mm0
# CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
# CHECK: [0] [1] [2] [3]
-# CHECK-NEXT: 0. 2 2.0 0.5 0.0 pcmpeqw %mm0, %mm0
-# CHECK-NEXT: 1. 2 3.0 0.0 0.0 paddw %mm0, %mm0
-# CHECK-NEXT: 2 2.5 0.3 0.0 <total>
+# CHECK-NEXT: 0. 2 1.0 1.0 0.5 pcmpeqw %mm0, %mm0
+# CHECK-NEXT: 1. 2 2.0 0.0 0.0 paddw %mm0, %mm0
+# CHECK-NEXT: 2 1.5 0.5 0.3 <total>
# CHECK: [2] Code Region
# CHECK: Iterations: 1000
# CHECK-NEXT: Instructions: 2000
-# CHECK-NEXT: Total Cycles: 2003
+# CHECK-NEXT: Total Cycles: 504
# CHECK-NEXT: Total uOps: 2000
# CHECK: Dispatch Width: 6
-# CHECK-NEXT: uOps Per Cycle: 1.00
-# CHECK-NEXT: IPC: 1.00
+# CHECK-NEXT: uOps Per Cycle: 3.97
+# CHECK-NEXT: IPC: 3.97
# CHECK-NEXT: Block RThroughput: 0.5
# CHECK: Instruction Info:
@@ -224,12 +224,12 @@ paddd %mm0, %mm0
# CHECK: Register File statistics:
# CHECK-NEXT: Total number of mappings created: 2000
-# CHECK-NEXT: Max number of mappings used: 66
+# CHECK-NEXT: Max number of mappings used: 74
# CHECK: * Register File #1 -- Zn3FpPRF:
# CHECK-NEXT: Number of physical registers: 160
# CHECK-NEXT: Total number of mappings created: 2000
-# CHECK-NEXT: Max number of mappings used: 66
+# CHECK-NEXT: Max number of mappings used: 74
# CHECK: * Register File #2 -- Zn3IntegerPRF:
# CHECK-NEXT: Number of physical registers: 192
@@ -267,16 +267,16 @@ paddd %mm0, %mm0
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12.0] [12.1] [13] [14.0] [14.1] [14.2] [15.0] [15.1] [15.2] [16.0] [16.1] Instructions:
-# CHECK-NEXT: - - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - pcmpeqd %mm0, %mm0
-# CHECK-NEXT: - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - - paddd %mm0, %mm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 0.50 - - - - - - - - - - - - pcmpeqd %mm0, %mm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 - 0.50 - - - - - - - - - - - paddd %mm0, %mm0
# CHECK: Timeline view:
-# CHECK-NEXT: Index 0123456
+# CHECK-NEXT: Index 01234
-# CHECK: [0,0] DeER .. pcmpeqd %mm0, %mm0
-# CHECK-NEXT: [0,1] D=eER.. paddd %mm0, %mm0
-# CHECK-NEXT: [1,0] D==eER. pcmpeqd %mm0, %mm0
-# CHECK-NEXT: [1,1] D===eER paddd %mm0, %mm0
+# CHECK: [0,0] DeER. pcmpeqd %mm0, %mm0
+# CHECK-NEXT: [0,1] D=eER paddd %mm0, %mm0
+# CHECK-NEXT: [1,0] DeE-R pcmpeqd %mm0, %mm0
+# CHECK-NEXT: [1,1] D=eER paddd %mm0, %mm0
# CHECK: Average Wait times (based on the timeline view):
# CHECK-NEXT: [0]: Executions
@@ -285,6 +285,6 @@ paddd %mm0, %mm0
# CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
# CHECK: [0] [1] [2] [3]
-# CHECK-NEXT: 0. 2 2.0 0.5 0.0 pcmpeqd %mm0, %mm0
-# CHECK-NEXT: 1. 2 3.0 0.0 0.0 paddd %mm0, %mm0
-# CHECK-NEXT: 2 2.5 0.3 0.0 <total>
+# CHECK-NEXT: 0. 2 1.0 1.0 0.5 pcmpeqd %mm0, %mm0
+# CHECK-NEXT: 1. 2 2.0 0.0 0.0 paddd %mm0, %mm0
+# CHECK-NEXT: 2 1.5 0.5 0.3 <total>
More information about the llvm-commits
mailing list