From: Roman Lebedev Date: Mon, 10 May 2021 19:52:15 +0000 (+0300) Subject: [X86] AMD Zen 3: same-reg PCMPEQ is an MMX all-ones dep breaking idiom X-Git-Tag: llvmorg-14-init~7120 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=b24edfff4fb16549b3e5ec434ca79dd86fdb4e43;p=platform%2Fupstream%2Fllvm.git [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. --- diff --git a/llvm/lib/Target/X86/X86ScheduleZnver3.td b/llvm/lib/Target/X86/X86ScheduleZnver3.td index f377fb9..43a23c8 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 f339f69..5909af8 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 +# 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 # 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 +# 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 # 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 +# 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