[llvm] 0f3bcb9 - [X86] AMD Zen 3: same-reg SSE XMM PCMP is dep breaking one-idiom
Roman Lebedev via llvm-commits
llvm-commits at lists.llvm.org
Mon May 10 13:50:00 PDT 2021
Author: Roman Lebedev
Date: 2021-05-10T23:49:26+03:00
New Revision: 0f3bcb97efa8ac6c3277390c3fa2085ee72b074e
URL: https://github.com/llvm/llvm-project/commit/0f3bcb97efa8ac6c3277390c3fa2085ee72b074e
DIFF: https://github.com/llvm/llvm-project/commit/0f3bcb97efa8ac6c3277390c3fa2085ee72b074e.diff
LOG: [X86] AMD Zen 3: same-reg SSE XMM PCMP is dep breaking one-idiom
As measured by exegesis, and confirmed by ref docs.
Much like with MMX PCMP, it does actually have to execute, though.
Added:
Modified:
llvm/lib/Target/X86/X86ScheduleZnver3.td
llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-sse-xmm.s
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86ScheduleZnver3.td b/llvm/lib/Target/X86/X86ScheduleZnver3.td
index 43a23c8ab88d..b1e1e4cc8ab7 100644
--- a/llvm/lib/Target/X86/X86ScheduleZnver3.td
+++ b/llvm/lib/Target/X86/X86ScheduleZnver3.td
@@ -1551,6 +1551,11 @@ def : IsDepBreakingFunction<[
DepBreakingClass<[
MMX_PCMPEQBirr, MMX_PCMPEQWirr, MMX_PCMPEQDirr
], ZeroIdiomPredicate>,
+
+ // SSE
+ DepBreakingClass<[
+ PCMPEQBrr, PCMPEQWrr, PCMPEQDrr, PCMPEQQrr
+ ], ZeroIdiomPredicate>,
]>;
} // SchedModel
diff --git a/llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-sse-xmm.s b/llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-sse-xmm.s
index 2f483f242d5b..5a05487f217a 100644
--- a/llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-sse-xmm.s
+++ b/llvm/test/tools/llvm-mca/X86/Znver3/one-idioms-sse-xmm.s
@@ -25,12 +25,12 @@ paddq %xmm0, %xmm0
# 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:
@@ -47,12 +47,12 @@ paddq %xmm0, %xmm0
# 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
@@ -90,16 +90,16 @@ paddq %xmm0, %xmm0
# 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 %xmm0, %xmm0
-# CHECK-NEXT: - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - - paddb %xmm0, %xmm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 0.50 - - - - - - - - - - - - pcmpeqb %xmm0, %xmm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 - 0.50 - - - - - - - - - - - paddb %xmm0, %xmm0
# CHECK: Timeline view:
-# CHECK-NEXT: Index 0123456
+# CHECK-NEXT: Index 01234
-# CHECK: [0,0] DeER .. pcmpeqb %xmm0, %xmm0
-# CHECK-NEXT: [0,1] D=eER.. paddb %xmm0, %xmm0
-# CHECK-NEXT: [1,0] D==eER. pcmpeqb %xmm0, %xmm0
-# CHECK-NEXT: [1,1] D===eER paddb %xmm0, %xmm0
+# CHECK: [0,0] DeER. pcmpeqb %xmm0, %xmm0
+# CHECK-NEXT: [0,1] D=eER paddb %xmm0, %xmm0
+# CHECK-NEXT: [1,0] DeE-R pcmpeqb %xmm0, %xmm0
+# CHECK-NEXT: [1,1] D=eER paddb %xmm0, %xmm0
# CHECK: Average Wait times (based on the timeline view):
# CHECK-NEXT: [0]: Executions
@@ -108,20 +108,20 @@ paddq %xmm0, %xmm0
# 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 %xmm0, %xmm0
-# CHECK-NEXT: 1. 2 3.0 0.0 0.0 paddb %xmm0, %xmm0
-# CHECK-NEXT: 2 2.5 0.3 0.0 <total>
+# CHECK-NEXT: 0. 2 1.0 1.0 0.5 pcmpeqb %xmm0, %xmm0
+# CHECK-NEXT: 1. 2 2.0 0.0 0.0 paddb %xmm0, %xmm0
+# 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:
@@ -138,12 +138,12 @@ paddq %xmm0, %xmm0
# 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
@@ -181,16 +181,16 @@ paddq %xmm0, %xmm0
# 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 %xmm0, %xmm0
-# CHECK-NEXT: - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - - paddw %xmm0, %xmm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 0.50 - - - - - - - - - - - - pcmpeqw %xmm0, %xmm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 - 0.50 - - - - - - - - - - - paddw %xmm0, %xmm0
# CHECK: Timeline view:
-# CHECK-NEXT: Index 0123456
+# CHECK-NEXT: Index 01234
-# CHECK: [0,0] DeER .. pcmpeqw %xmm0, %xmm0
-# CHECK-NEXT: [0,1] D=eER.. paddw %xmm0, %xmm0
-# CHECK-NEXT: [1,0] D==eER. pcmpeqw %xmm0, %xmm0
-# CHECK-NEXT: [1,1] D===eER paddw %xmm0, %xmm0
+# CHECK: [0,0] DeER. pcmpeqw %xmm0, %xmm0
+# CHECK-NEXT: [0,1] D=eER paddw %xmm0, %xmm0
+# CHECK-NEXT: [1,0] DeE-R pcmpeqw %xmm0, %xmm0
+# CHECK-NEXT: [1,1] D=eER paddw %xmm0, %xmm0
# CHECK: Average Wait times (based on the timeline view):
# CHECK-NEXT: [0]: Executions
@@ -199,20 +199,20 @@ paddq %xmm0, %xmm0
# 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 %xmm0, %xmm0
-# CHECK-NEXT: 1. 2 3.0 0.0 0.0 paddw %xmm0, %xmm0
-# CHECK-NEXT: 2 2.5 0.3 0.0 <total>
+# CHECK-NEXT: 0. 2 1.0 1.0 0.5 pcmpeqw %xmm0, %xmm0
+# CHECK-NEXT: 1. 2 2.0 0.0 0.0 paddw %xmm0, %xmm0
+# 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:
@@ -229,12 +229,12 @@ paddq %xmm0, %xmm0
# 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
@@ -272,16 +272,16 @@ paddq %xmm0, %xmm0
# 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 %xmm0, %xmm0
-# CHECK-NEXT: - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - - paddd %xmm0, %xmm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 0.50 - - - - - - - - - - - - pcmpeqd %xmm0, %xmm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 - 0.50 - - - - - - - - - - - paddd %xmm0, %xmm0
# CHECK: Timeline view:
-# CHECK-NEXT: Index 0123456
+# CHECK-NEXT: Index 01234
-# CHECK: [0,0] DeER .. pcmpeqd %xmm0, %xmm0
-# CHECK-NEXT: [0,1] D=eER.. paddd %xmm0, %xmm0
-# CHECK-NEXT: [1,0] D==eER. pcmpeqd %xmm0, %xmm0
-# CHECK-NEXT: [1,1] D===eER paddd %xmm0, %xmm0
+# CHECK: [0,0] DeER. pcmpeqd %xmm0, %xmm0
+# CHECK-NEXT: [0,1] D=eER paddd %xmm0, %xmm0
+# CHECK-NEXT: [1,0] DeE-R pcmpeqd %xmm0, %xmm0
+# CHECK-NEXT: [1,1] D=eER paddd %xmm0, %xmm0
# CHECK: Average Wait times (based on the timeline view):
# CHECK-NEXT: [0]: Executions
@@ -290,20 +290,20 @@ paddq %xmm0, %xmm0
# 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 %xmm0, %xmm0
-# CHECK-NEXT: 1. 2 3.0 0.0 0.0 paddd %xmm0, %xmm0
-# CHECK-NEXT: 2 2.5 0.3 0.0 <total>
+# CHECK-NEXT: 0. 2 1.0 1.0 0.5 pcmpeqd %xmm0, %xmm0
+# CHECK-NEXT: 1. 2 2.0 0.0 0.0 paddd %xmm0, %xmm0
+# CHECK-NEXT: 2 1.5 0.5 0.3 <total>
# CHECK: [3] 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:
@@ -320,12 +320,12 @@ paddq %xmm0, %xmm0
# 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
@@ -363,16 +363,16 @@ paddq %xmm0, %xmm0
# 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 - - - - - - - - - - - pcmpeqq %xmm0, %xmm0
-# CHECK-NEXT: - - - - - - - - 0.50 - 0.50 - - - - - - - - - - - - paddq %xmm0, %xmm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 0.50 - - - - - - - - - - - - pcmpeqq %xmm0, %xmm0
+# CHECK-NEXT: - - - - - - - - 0.25 0.25 - 0.50 - - - - - - - - - - - paddq %xmm0, %xmm0
# CHECK: Timeline view:
-# CHECK-NEXT: Index 0123456
+# CHECK-NEXT: Index 01234
-# CHECK: [0,0] DeER .. pcmpeqq %xmm0, %xmm0
-# CHECK-NEXT: [0,1] D=eER.. paddq %xmm0, %xmm0
-# CHECK-NEXT: [1,0] D==eER. pcmpeqq %xmm0, %xmm0
-# CHECK-NEXT: [1,1] D===eER paddq %xmm0, %xmm0
+# CHECK: [0,0] DeER. pcmpeqq %xmm0, %xmm0
+# CHECK-NEXT: [0,1] D=eER paddq %xmm0, %xmm0
+# CHECK-NEXT: [1,0] DeE-R pcmpeqq %xmm0, %xmm0
+# CHECK-NEXT: [1,1] D=eER paddq %xmm0, %xmm0
# CHECK: Average Wait times (based on the timeline view):
# CHECK-NEXT: [0]: Executions
@@ -381,6 +381,6 @@ paddq %xmm0, %xmm0
# 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 pcmpeqq %xmm0, %xmm0
-# CHECK-NEXT: 1. 2 3.0 0.0 0.0 paddq %xmm0, %xmm0
-# CHECK-NEXT: 2 2.5 0.3 0.0 <total>
+# CHECK-NEXT: 0. 2 1.0 1.0 0.5 pcmpeqq %xmm0, %xmm0
+# CHECK-NEXT: 1. 2 2.0 0.0 0.0 paddq %xmm0, %xmm0
+# CHECK-NEXT: 2 1.5 0.5 0.3 <total>
More information about the llvm-commits
mailing list