Matthew Poremba has uploaded this change for review. ( https://gem5-review.googlesource.com/c/public/gem5/+/67074?usp=email )

Change subject: arch-vega: Implement ds_add_f32 atomic
......................................................................

arch-vega: Implement ds_add_f32 atomic

This instruction does an atomic add of a 32-bit float with a VGPR and
value in LDS atomically without return.

Change-Id: Id4f23a1ab587a23edfd1d88ede1cbcc5bdedc0cb
---
M src/arch/amdgpu/vega/insts/instructions.cc
M src/arch/amdgpu/vega/insts/instructions.hh
2 files changed, 60 insertions(+), 3 deletions(-)



diff --git a/src/arch/amdgpu/vega/insts/instructions.cc b/src/arch/amdgpu/vega/insts/instructions.cc
index 5332687..a0308c8 100644
--- a/src/arch/amdgpu/vega/insts/instructions.cc
+++ b/src/arch/amdgpu/vega/insts/instructions.cc
@@ -34749,6 +34749,10 @@
         : Inst_DS(iFmt, "ds_add_f32")
     {
         setFlag(F32);
+        setFlag(MemoryRef);
+        setFlag(GroupSegment);
+        setFlag(AtomicAdd);
+        setFlag(AtomicNoReturn);
     } // Inst_DS__DS_ADD_F32

     Inst_DS__DS_ADD_F32::~Inst_DS__DS_ADD_F32()
@@ -34757,15 +34761,54 @@

     // --- description from .arch file ---
     // 32b:
-    // tmp = MEM[ADDR];
     // MEM[ADDR] += DATA;
-    // RETURN_DATA = tmp.
     // Floating point add that handles NaN/INF/denormal values.
     void
     Inst_DS__DS_ADD_F32::execute(GPUDynInstPtr gpuDynInst)
     {
-        panicUnimplemented();
+        Wavefront *wf = gpuDynInst->wavefront();
+
+        if (gpuDynInst->exec_mask.none()) {
+            wf->decLGKMInstsIssued();
+            return;
+        }
+
+        gpuDynInst->execUnitId = wf->execUnitId;
+        gpuDynInst->latency.init(gpuDynInst->computeUnit());
+        gpuDynInst->latency.set(
+                gpuDynInst->computeUnit()->cyclesToTicks(Cycles(24)));
+        ConstVecOperandU32 addr(gpuDynInst, extData.ADDR);
+        ConstVecOperandF32 data(gpuDynInst, extData.DATA0);
+
+        addr.read();
+        data.read();
+
+        calcAddr(gpuDynInst, addr);
+
+        for (int lane = 0; lane < NumVecElemPerVecReg; ++lane) {
+            if (gpuDynInst->exec_mask[lane]) {
+                (reinterpret_cast<VecElemF32*>(gpuDynInst->a_data))[lane]
+                    = data[lane];
+            }
+        }
+
+ gpuDynInst->computeUnit()->localMemoryPipe.issueRequest(gpuDynInst);
     } // execute
+
+    void
+    Inst_DS__DS_ADD_F32::initiateAcc(GPUDynInstPtr gpuDynInst)
+    {
+        Addr offset0 = instData.OFFSET0;
+        Addr offset1 = instData.OFFSET1;
+        Addr offset = (offset1 << 8) | offset0;
+
+        initAtomicAccess<VecElemF32>(gpuDynInst, offset);
+    } // initiateAcc
+
+    void
+    Inst_DS__DS_ADD_F32::completeAcc(GPUDynInstPtr gpuDynInst)
+    {
+    } // completeAcc
     // --- Inst_DS__DS_WRITE_B8 class methods ---

     Inst_DS__DS_WRITE_B8::Inst_DS__DS_WRITE_B8(InFmt_DS *iFmt)
diff --git a/src/arch/amdgpu/vega/insts/instructions.hh b/src/arch/amdgpu/vega/insts/instructions.hh
index 33be33e..05a0002 100644
--- a/src/arch/amdgpu/vega/insts/instructions.hh
+++ b/src/arch/amdgpu/vega/insts/instructions.hh
@@ -31895,6 +31895,8 @@
             }
         } // getOperandSize

+        void initiateAcc(GPUDynInstPtr gpuDynInst) override;
+        void completeAcc(GPUDynInstPtr gpuDynInst) override;
         void execute(GPUDynInstPtr) override;
     }; // Inst_DS__DS_ADD_F32


--
To view, visit https://gem5-review.googlesource.com/c/public/gem5/+/67074?usp=email To unsubscribe, or for help writing mail filters, visit https://gem5-review.googlesource.com/settings

Gerrit-Project: public/gem5
Gerrit-Branch: develop
Gerrit-Change-Id: Id4f23a1ab587a23edfd1d88ede1cbcc5bdedc0cb
Gerrit-Change-Number: 67074
Gerrit-PatchSet: 1
Gerrit-Owner: Matthew Poremba <matthew.pore...@amd.com>
Gerrit-MessageType: newchange
_______________________________________________
gem5-dev mailing list -- gem5-dev@gem5.org
To unsubscribe send an email to gem5-dev-le...@gem5.org

Reply via email to