[RISCV] Split PseudoVFMUL by SEW
Co-authored-by: Wang Pengcheng <wangpengcheng.pp@bytedance.com>
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
index 1097eb4..c00c650 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
@@ -2832,16 +2832,17 @@
multiclass VPseudoVFMUL_VV_VF_RM {
foreach m = MxListF in {
- defm "" : VPseudoBinaryFV_VV_RM<m>,
- SchedBinary<"WriteVFMulV", "ReadVFMulV", "ReadVFMulV", m.MX,
- forceMergeOpRead=true>;
+ foreach e = SchedSEWSet<m.MX, isF=1>.val in
+ defm "" : VPseudoBinaryFV_VV_RM<m, "", sew=e>,
+ SchedBinary<"WriteVFMulV", "ReadVFMulV", "ReadVFMulV", m.MX, e,
+ forceMergeOpRead=true>;
}
foreach f = FPList in {
foreach m = f.MxList in {
- defm "" : VPseudoBinaryV_VF_RM<m, f>,
+ defm "" : VPseudoBinaryV_VF_RM<m, f, "", sew=f.SEW>,
SchedBinary<"WriteVFMulF", "ReadVFMulV", "ReadVFMulF", m.MX,
- forceMergeOpRead=true>;
+ f.SEW, forceMergeOpRead=true>;
}
}
}
@@ -7106,7 +7107,7 @@
// 13.4. Vector Single-Width Floating-Point Multiply/Divide Instructions
//===----------------------------------------------------------------------===//
defm : VPatBinaryV_VV_VX_RM<"int_riscv_vfmul", "PseudoVFMUL",
- AllFloatVectors>;
+ AllFloatVectors, isSEWAware=1>;
defm : VPatBinaryV_VV_VX_RM<"int_riscv_vfdiv", "PseudoVFDIV",
AllFloatVectors, isSEWAware=1>;
defm : VPatBinaryV_VX_RM<"int_riscv_vfrdiv", "PseudoVFRDIV",
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
index 37acd96..26847e1 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
@@ -1213,7 +1213,7 @@
defm : VPatWidenBinaryFPSDNode_VV_VF_WV_WF_RM<fsub, "PseudoVFWSUB">;
// 13.4. Vector Single-Width Floating-Point Multiply/Divide Instructions
-defm : VPatBinaryFPSDNode_VV_VF_RM<any_fmul, "PseudoVFMUL">;
+defm : VPatBinaryFPSDNode_VV_VF_RM<any_fmul, "PseudoVFMUL", isSEWAware=1>;
defm : VPatBinaryFPSDNode_VV_VF_RM<any_fdiv, "PseudoVFDIV", isSEWAware=1>;
defm : VPatBinaryFPSDNode_R_VF_RM<any_fdiv, "PseudoVFRDIV", isSEWAware=1>;
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
index d2f5754..3225260 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
@@ -2443,7 +2443,7 @@
"PseudoVFWSUB", isSEWAware=1>;
// 13.4. Vector Single-Width Floating-Point Multiply/Divide Instructions
-defm : VPatBinaryFPVL_VV_VF_RM<any_riscv_fmul_vl, "PseudoVFMUL">;
+defm : VPatBinaryFPVL_VV_VF_RM<any_riscv_fmul_vl, "PseudoVFMUL", isSEWAware=1>;
defm : VPatBinaryFPVL_VV_VF_RM<any_riscv_fdiv_vl, "PseudoVFDIV", isSEWAware=1>;
defm : VPatBinaryFPVL_R_VF_RM<any_riscv_fdiv_vl, "PseudoVFRDIV", isSEWAware=1>;
diff --git a/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td b/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td
index 2b6fc5e..af65b70 100644
--- a/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td
+++ b/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td
@@ -739,6 +739,8 @@
let Latency = 8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, Cycles)] in {
defm "" : LMULSEWWriteResMXSEW<"WriteVFALUV", [SiFive7VCQ, SiFive7VA], mx, sew, IsWorstCase>;
defm "" : LMULSEWWriteResMXSEW<"WriteVFALUF", [SiFive7VCQ, SiFive7VA], mx, sew, IsWorstCase>;
+ defm "" : LMULSEWWriteResMXSEW<"WriteVFMulV", [SiFive7VCQ, SiFive7VA], mx, sew, IsWorstCase>;
+ defm "" : LMULSEWWriteResMXSEW<"WriteVFMulF", [SiFive7VCQ, SiFive7VA], mx, sew, IsWorstCase>;
}
}
}
@@ -746,8 +748,6 @@
defvar Cycles = SiFive7GetCyclesDefault<mx>.c;
defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxList>.c;
let Latency = 8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, Cycles)] in {
- defm "" : LMULWriteResMX<"WriteVFMulV", [SiFive7VCQ, SiFive7VA], mx, IsWorstCase>;
- defm "" : LMULWriteResMX<"WriteVFMulF", [SiFive7VCQ, SiFive7VA], mx, IsWorstCase>;
defm "" : LMULWriteResMX<"WriteVFMulAddV", [SiFive7VCQ, SiFive7VA], mx, IsWorstCase>;
defm "" : LMULWriteResMX<"WriteVFMulAddF", [SiFive7VCQ, SiFive7VA], mx, IsWorstCase>;
defm "" : LMULWriteResMX<"WriteVFRecpV", [SiFive7VCQ, SiFive7VA], mx, IsWorstCase>;
@@ -1157,8 +1157,8 @@
defm "" : LMULSEWReadAdvanceF<"ReadVFALUF", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWALUV", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWALUF", 0>;
-defm "" : LMULReadAdvance<"ReadVFMulV", 0>;
-defm "" : LMULReadAdvance<"ReadVFMulF", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFMulV", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFMulF", 0>;
defm "" : LMULSEWReadAdvanceF<"ReadVFDivV", 0>;
defm "" : LMULSEWReadAdvanceF<"ReadVFDivF", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWMulV", 0>;
diff --git a/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td b/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td
index 00e92cd..5819ce5 100644
--- a/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td
+++ b/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td
@@ -493,6 +493,9 @@
let Latency = 6, ReleaseAtCycles = [LMulLat] in {
defm "" : LMULSEWWriteResMXSEW<"WriteVFALUV", [SiFiveP600VectorArith], mx, sew, IsWorstCase>;
defm "" : LMULSEWWriteResMXSEW<"WriteVFALUF", [SiFiveP600VectorArith], mx, sew, IsWorstCase>;
+ defm "" : LMULSEWWriteResMXSEW<"WriteVFMulV", [SiFiveP600VectorArith], mx, sew, IsWorstCase>;
+ defm "" : LMULSEWWriteResMXSEW<"WriteVFMulF", [SiFiveP600VectorArith], mx, sew, IsWorstCase>;
+
}
}
}
@@ -500,8 +503,6 @@
defvar LMulLat = SiFiveP600GetLMulCycles<mx>.c;
defvar IsWorstCase = SiFiveP600IsWorstCaseMX<mx, SchedMxList>.c;
let Latency = 6, ReleaseAtCycles = [LMulLat] in {
- defm "" : LMULWriteResMX<"WriteVFMulV", [SiFiveP600VectorArith], mx, IsWorstCase>;
- defm "" : LMULWriteResMX<"WriteVFMulF", [SiFiveP600VectorArith], mx, IsWorstCase>;
defm "" : LMULWriteResMX<"WriteVFMulAddV", [SiFiveP600VectorArith], mx, IsWorstCase>;
defm "" : LMULWriteResMX<"WriteVFMulAddF", [SiFiveP600VectorArith], mx, IsWorstCase>;
}
@@ -945,8 +946,8 @@
defm "" : LMULSEWReadAdvanceF<"ReadVFALUF", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWALUV", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWALUF", 0>;
-defm "" : LMULReadAdvance<"ReadVFMulV", 0>;
-defm "" : LMULReadAdvance<"ReadVFMulF", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFMulV", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFMulF", 0>;
defm "" : LMULSEWReadAdvanceF<"ReadVFDivV", 0>;
defm "" : LMULSEWReadAdvanceF<"ReadVFDivF", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWMulV", 0>;
diff --git a/llvm/lib/Target/RISCV/RISCVScheduleV.td b/llvm/lib/Target/RISCV/RISCVScheduleV.td
index 8b5bd7a..9cb3cd1 100644
--- a/llvm/lib/Target/RISCV/RISCVScheduleV.td
+++ b/llvm/lib/Target/RISCV/RISCVScheduleV.td
@@ -415,8 +415,8 @@
defm "" : LMULSEWSchedWritesFW<"WriteVFWALUV">;
defm "" : LMULSEWSchedWritesFW<"WriteVFWALUF">;
// 13.4. Vector Single-Width Floating-Point Multiply/Divide Instructions
-defm "" : LMULSchedWrites<"WriteVFMulV">;
-defm "" : LMULSchedWrites<"WriteVFMulF">;
+defm "" : LMULSEWSchedWritesF<"WriteVFMulV">;
+defm "" : LMULSEWSchedWritesF<"WriteVFMulF">;
defm "" : LMULSEWSchedWritesF<"WriteVFDivV">;
defm "" : LMULSEWSchedWritesF<"WriteVFDivF">;
// 13.5. Vector Widening Floating-Point Multiply
@@ -640,8 +640,8 @@
defm "" : LMULSEWSchedReadsFW<"ReadVFWALUV">;
defm "" : LMULSEWSchedReadsFW<"ReadVFWALUF">;
// 13.4. Vector Single-Width Floating-Point Multiply/Divide Instructions
-defm "" : LMULSchedReads<"ReadVFMulV">;
-defm "" : LMULSchedReads<"ReadVFMulF">;
+defm "" : LMULSEWSchedReadsF<"ReadVFMulV">;
+defm "" : LMULSEWSchedReadsF<"ReadVFMulF">;
defm "" : LMULSEWSchedReadsF<"ReadVFDivV">;
defm "" : LMULSEWSchedReadsF<"ReadVFDivF">;
// 13.5. Vector Widening Floating-Point Multiply
@@ -884,8 +884,8 @@
defm "" : LMULSEWWriteResF<"WriteVFALUF", []>;
defm "" : LMULSEWWriteResFW<"WriteVFWALUV", []>;
defm "" : LMULSEWWriteResFW<"WriteVFWALUF", []>;
-defm "" : LMULWriteRes<"WriteVFMulV", []>;
-defm "" : LMULWriteRes<"WriteVFMulF", []>;
+defm "" : LMULSEWWriteResF<"WriteVFMulV", []>;
+defm "" : LMULSEWWriteResF<"WriteVFMulF", []>;
defm "" : LMULSEWWriteResF<"WriteVFDivV", []>;
defm "" : LMULSEWWriteResF<"WriteVFDivF", []>;
defm "" : LMULSEWWriteResFW<"WriteVFWMulV", []>;
@@ -1040,8 +1040,8 @@
defm "" : LMULSEWReadAdvanceF<"ReadVFALUF", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWALUV", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWALUF", 0>;
-defm "" : LMULReadAdvance<"ReadVFMulV", 0>;
-defm "" : LMULReadAdvance<"ReadVFMulF", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFMulV", 0>;
+defm "" : LMULSEWReadAdvanceF<"ReadVFMulF", 0>;
defm "" : LMULSEWReadAdvanceF<"ReadVFDivV", 0>;
defm "" : LMULSEWReadAdvanceF<"ReadVFDivF", 0>;
defm "" : LMULSEWReadAdvanceFW<"ReadVFWMulV", 0>;
