diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp --- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp +++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp @@ -1884,6 +1884,9 @@ int GCNHazardRecognizer::checkFPAtomicToDenormModeHazard(MachineInstr *MI) { int FPAtomicToDenormModeWaitStates = 3; + if (!ST.hasFPAtomicToDenormModeHazard()) + return 0; + if (MI->getOpcode() != AMDGPU::S_DENORM_MODE) return 0; diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h --- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h +++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h @@ -1040,6 +1040,10 @@ bool hasGFX90AInsts() const { return GFX90AInsts; } + bool hasFPAtomicToDenormModeHazard() const { + return getGeneration() == GFX10; + } + bool hasVOP3DPP() const { return getGeneration() >= GFX11; } bool hasLdsDirect() const { return getGeneration() >= GFX11; } diff --git a/llvm/test/CodeGen/AMDGPU/fp-atomic-to-s_denormmode.mir b/llvm/test/CodeGen/AMDGPU/fp-atomic-to-s_denormmode.mir --- a/llvm/test/CodeGen/AMDGPU/fp-atomic-to-s_denormmode.mir +++ b/llvm/test/CodeGen/AMDGPU/fp-atomic-to-s_denormmode.mir @@ -1,9 +1,9 @@ -# RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -run-pass post-RA-hazard-rec -o - %s | FileCheck -check-prefix=GCN %s -# RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -run-pass post-RA-hazard-rec -o - %s | FileCheck -check-prefix=GCN %s +# RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -run-pass post-RA-hazard-rec -o - %s | FileCheck -check-prefixes=GCN,GFX10 %s +# RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -run-pass post-RA-hazard-rec -o - %s | FileCheck -check-prefixes=GCN %s # GCN-LABEL: name: flat_atomic_fcmpswap_to_s_denorm_mode # GCN: FLAT_ATOMIC_FCMPSWAP -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fcmpswap_to_s_denorm_mode @@ -15,7 +15,7 @@ # GCN-LABEL: name: flat_atomic_fcmpswap_x2_to_s_denorm_mode # GCN: FLAT_ATOMIC_FCMPSWAP_X2 -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fcmpswap_x2_to_s_denorm_mode @@ -27,7 +27,7 @@ # GCN-LABEL: name: flat_atomic_fmax_to_s_denorm_mode # GCN: FLAT_ATOMIC_FMAX -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fmax_to_s_denorm_mode @@ -39,7 +39,7 @@ # GCN-LABEL: name: flat_atomic_fmax_x2_to_s_denorm_mode # GCN: FLAT_ATOMIC_FMAX_X2 -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fmax_x2_to_s_denorm_mode @@ -51,7 +51,7 @@ # GCN-LABEL: name: flat_atomic_fmin_to_s_denorm_mode # GCN: FLAT_ATOMIC_FMIN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fmin_to_s_denorm_mode @@ -63,7 +63,7 @@ # GCN-LABEL: name: flat_atomic_fmin_x2_to_s_denorm_mode # GCN: FLAT_ATOMIC_FMIN_X2 -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fmin_x2_to_s_denorm_mode @@ -75,7 +75,7 @@ # GCN-LABEL: name: flat_atomic_fcmpswap_x2_rtn_to_s_denorm_mode # GCN: FLAT_ATOMIC_FCMPSWAP_X2_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fcmpswap_x2_rtn_to_s_denorm_mode @@ -87,7 +87,7 @@ # GCN-LABEL: name: flat_atomic_fmax_rtn_to_s_denorm_mode # GCN: FLAT_ATOMIC_FMAX_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fmax_rtn_to_s_denorm_mode @@ -99,7 +99,7 @@ # GCN-LABEL: name: flat_atomic_fmax_x2_rtn_to_s_denorm_mode # GCN: FLAT_ATOMIC_FMAX_X2_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fmax_x2_rtn_to_s_denorm_mode @@ -111,7 +111,7 @@ # GCN-LABEL: name: flat_atomic_fmin_rtn_to_s_denorm_mode # GCN: FLAT_ATOMIC_FMIN_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fmin_rtn_to_s_denorm_mode @@ -123,7 +123,7 @@ # GCN-LABEL: name: flat_atomic_fmin_x2_rtn_to_s_denorm_mode # GCN: FLAT_ATOMIC_FMIN_X2_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fmin_x2_rtn_to_s_denorm_mode @@ -135,7 +135,7 @@ # GCN-LABEL: name: flat_atomic_fcmpswap_rtn_to_s_denorm_mode # GCN: FLAT_ATOMIC_FCMPSWAP_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: flat_atomic_fcmpswap_rtn_to_s_denorm_mode @@ -147,7 +147,7 @@ # GCN-LABEL: name: global_atomic_fcmpswap_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FCMPSWAP -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fcmpswap_to_s_denorm_mode @@ -159,7 +159,7 @@ # GCN-LABEL: name: global_atomic_fcmpswap_x2_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FCMPSWAP_X2 -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fcmpswap_x2_to_s_denorm_mode @@ -171,7 +171,7 @@ # GCN-LABEL: name: global_atomic_fmax_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMAX -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmax_to_s_denorm_mode @@ -183,7 +183,7 @@ # GCN-LABEL: name: global_atomic_fmax_x2_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMAX_X2 -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmax_x2_to_s_denorm_mode @@ -195,7 +195,7 @@ # GCN-LABEL: name: global_atomic_fmin_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMIN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmin_to_s_denorm_mode @@ -207,7 +207,7 @@ # GCN-LABEL: name: global_atomic_fmin_x2_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMIN_X2 -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmin_x2_to_s_denorm_mode @@ -219,7 +219,7 @@ # GCN-LABEL: name: global_atomic_fcmpswap_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FCMPSWAP_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fcmpswap_rtn_to_s_denorm_mode @@ -231,7 +231,7 @@ # GCN-LABEL: name: global_atomic_fcmpswap_x2_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FCMPSWAP_X2_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fcmpswap_x2_rtn_to_s_denorm_mode @@ -243,7 +243,7 @@ # GCN-LABEL: name: global_atomic_fmax_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMAX_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmax_rtn_to_s_denorm_mode @@ -255,7 +255,7 @@ # GCN-LABEL: name: global_atomic_fmax_x2_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMAX_X2_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmax_x2_rtn_to_s_denorm_mode @@ -267,7 +267,7 @@ # GCN-LABEL: name: global_atomic_fmin_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMIN_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmin_rtn_to_s_denorm_mode @@ -279,7 +279,7 @@ # GCN-LABEL: name: global_atomic_fmin_x2_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMIN_X2_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmin_x2_rtn_to_s_denorm_mode @@ -291,7 +291,7 @@ # GCN-LABEL: name: global_atomic_fcmpswap_saddr_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FCMPSWAP_SADDR -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fcmpswap_saddr_to_s_denorm_mode @@ -303,7 +303,7 @@ # GCN-LABEL: name: global_atomic_fcmpswap_x2_saddr_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FCMPSWAP_X2_SADDR_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fcmpswap_x2_saddr_rtn_to_s_denorm_mode @@ -315,7 +315,7 @@ # GCN-LABEL: name: global_atomic_fmax_saddr_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMAX_SADDR_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmax_saddr_rtn_to_s_denorm_mode @@ -327,7 +327,7 @@ # GCN-LABEL: name: global_atomic_fmax_x2_saddr_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMAX_X2_SADDR_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmax_x2_saddr_rtn_to_s_denorm_mode @@ -339,7 +339,7 @@ # GCN-LABEL: name: global_atomic_fmin_saddr_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMIN_SADDR_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmin_saddr_rtn_to_s_denorm_mode @@ -351,7 +351,7 @@ # GCN-LABEL: name: global_atomic_fmin_x2_saddr_rtn_to_s_denorm_mode # GCN: GLOBAL_ATOMIC_FMIN_X2_SADDR_RTN -# GCN-NEXT: S_NOP 2 +# GFX10-NEXT: S_NOP 2 # GCN-NEXT: S_DENORM_MODE --- name: global_atomic_fmin_x2_saddr_rtn_to_s_denorm_mode