gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

.

Browse Source
This commit is contained in:
Philippe Tillet
2023-01-09 22:11:00 -08:00
parent d88353a5a4
commit ff04a5e9b6
4 changed files with 88 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
lib/Conversion/TritonGPUToLLVM/ConvertLayoutOpToLLVM.cpp
View File

@@ -72,24 +72,24 @@ void storeDistributedToShared(Value src, Value llSrc,
Value staIdx1 = i32_val(0); Value staIdx1 = i32_val(0);
Value stride0 = dstStrides[outOrd[0]]; Value stride0 = dstStrides[outOrd[0]];
Value stride1 = dstStrides[outOrd[1]]; Value stride1 = dstStrides[outOrd[1]];
// if (auto addOp = dyn_cast<LLVM::AddOp>(dynIdx0.getDefiningOp())) if (auto addOp = dyn_cast<LLVM::AddOp>(dynIdx0.getDefiningOp()))
// if (auto cstRhs = if (auto cstRhs =
// dyn_cast<LLVM::ConstantOp>(addOp.getRhs().getDefiningOp())) { dyn_cast<LLVM::ConstantOp>(addOp.getRhs().getDefiningOp())) {
// unsigned rhsVal = unsigned rhsVal =
// cstRhs.getValue().cast<IntegerAttr>().getValue().getSExtValue(); cstRhs.getValue().cast<IntegerAttr>().getValue().getSExtValue();
// unsigned key = (rhsVal / outVec) % maxPhase; unsigned key = (rhsVal / outVec) % maxPhase;
// if (cache.find(key) == cache.end()) if (cache.find(key) == cache.end())
// cache[key] = dynIdx0; cache[key] = dynIdx0;
// dynIdx0 = cache[key]; dynIdx0 = cache[key];
// staIdx0 = staIdx0 =
// i32_val((rhsVal) / (outVec * maxPhase) * (outVec * maxPhase)); i32_val((rhsVal) / (outVec * maxPhase) * (outVec * maxPhase));
// } }
// if (auto addOp = dyn_cast<LLVM::AddOp>(dynIdx1.getDefiningOp())) if (auto addOp = dyn_cast<LLVM::AddOp>(dynIdx1.getDefiningOp()))
// if (auto cstRhs = if (auto cstRhs =
// dyn_cast<LLVM::ConstantOp>(addOp.getRhs().getDefiningOp())) { dyn_cast<LLVM::ConstantOp>(addOp.getRhs().getDefiningOp())) {
// dynIdx1 = addOp.getLhs(); dynIdx1 = addOp.getLhs();
// staIdx1 = addOp.getRhs(); staIdx1 = addOp.getRhs();
// } }
// offset along non-contiguous dimension // offset along non-contiguous dimension
Value off1 = mul(dynIdx1, stride1); Value off1 = mul(dynIdx1, stride1);

80
lib/Dialect/TritonGPU/Transforms/Combine.cpp
View File

@@ -1266,14 +1266,16 @@ public:
: mlir::RewritePattern(triton::gpu::ConvertLayoutOp::getOperationName(), : mlir::RewritePattern(triton::gpu::ConvertLayoutOp::getOperationName(),
1, context) {} 1, context) {}
LogicalResult matchAndRewrite(mlir::Operation* op, LogicalResult
matchAndRewrite(mlir::Operation *op,
mlir::PatternRewriter &rewriter) const override { mlir::PatternRewriter &rewriter) const override {
auto dstOp = cast<triton::gpu::ConvertLayoutOp>(op); auto dstOp = cast<triton::gpu::ConvertLayoutOp>(op);
auto tmpOp = dyn_cast_or_null<triton::TransOp>(dstOp.src().getDefiningOp()); auto tmpOp = dyn_cast_or_null<triton::TransOp>(dstOp.src().getDefiningOp());
if(!tmpOp) if (!tmpOp)
return mlir::failure(); return mlir::failure();
auto srcOp = dyn_cast_or_null<triton::gpu::ConvertLayoutOp>(tmpOp.src().getDefiningOp()); auto srcOp = dyn_cast_or_null<triton::gpu::ConvertLayoutOp>(
if(!srcOp) tmpOp.src().getDefiningOp());
if (!srcOp)
return mlir::failure(); return mlir::failure();
auto arg = srcOp.src(); auto arg = srcOp.src();
auto X = tmpOp.src(); auto X = tmpOp.src();
@@ -1285,25 +1287,74 @@ public:
auto ZType = dstOp.getResult().getType().cast<RankedTensorType>(); auto ZType = dstOp.getResult().getType().cast<RankedTensorType>();
// encodings // encodings
auto argEncoding = argType.getEncoding(); auto argEncoding = argType.getEncoding();
auto XEncoding = XType.getEncoding().cast<triton::gpu::SharedEncodingAttr>(); auto XEncoding =
auto YEncoding = YType.getEncoding().cast<triton::gpu::SharedEncodingAttr>(); XType.getEncoding().cast<triton::gpu::SharedEncodingAttr>();
auto ZEncoding = ZType.getEncoding().dyn_cast<triton::gpu::DotOperandEncodingAttr>(); auto YEncoding =
if(!ZEncoding) YType.getEncoding().cast<triton::gpu::SharedEncodingAttr>();
auto ZEncoding =
ZType.getEncoding().dyn_cast<triton::gpu::DotOperandEncodingAttr>();
if (!ZEncoding)
return mlir::failure(); return mlir::failure();
// new X encoding // new X encoding
auto newXOrder = triton::gpu::getOrder(argEncoding); auto newXOrder = triton::gpu::getOrder(argEncoding);
auto newXEncoding = triton::gpu::SharedEncodingAttr::get( auto newXEncoding = triton::gpu::SharedEncodingAttr::get(
getContext(), ZEncoding, XType.getShape(), newXOrder, getContext(), ZEncoding, XType.getShape(), newXOrder,
XType.getElementType()); XType.getElementType());
auto newXType = RankedTensorType::get(XType.getShape(), XType.getElementType(), auto newXType = RankedTensorType::get(XType.getShape(),
newXEncoding); XType.getElementType(), newXEncoding);
if(XEncoding == newXEncoding) if (XEncoding == newXEncoding)
return mlir::failure(); return mlir::failure();
auto newX = rewriter.create<triton::gpu::ConvertLayoutOp>(srcOp.getLoc(),
auto newX = rewriter.create<triton::gpu::ConvertLayoutOp>(srcOp.getLoc(), newXType, arg); newXType, arg);
auto newY = rewriter.create<triton::TransOp>(tmpOp.getLoc(), newX); auto newY = rewriter.create<triton::TransOp>(tmpOp.getLoc(), newX);
rewriter.replaceOpWithNewOp<triton::gpu::ConvertLayoutOp>(dstOp, ZType, newY); rewriter.replaceOpWithNewOp<triton::gpu::ConvertLayoutOp>(dstOp, ZType,
newY);
return mlir::success();
}
};
//
class ConvertDotConvert : public mlir::RewritePattern {
public:
ConvertDotConvert(mlir::MLIRContext *context)
: mlir::RewritePattern(triton::gpu::ConvertLayoutOp::getOperationName(),
1, context) {}
LogicalResult
matchAndRewrite(mlir::Operation *op,
mlir::PatternRewriter &rewriter) const override {
auto dstOp = cast<triton::gpu::ConvertLayoutOp>(op);
auto dotOp = dyn_cast_or_null<triton::DotOp>(dstOp.src().getDefiningOp());
if (!dotOp)
return mlir::failure();
if (std::distance(dstOp->user_begin(), dstOp->user_end()) != 1 ||
std::distance(dotOp->user_begin(), dotOp->user_end()) != 1)
return mlir::failure();
auto cvtOp = dyn_cast_or_null<triton::gpu::ConvertLayoutOp>(
dotOp.getOperand(2).getDefiningOp());
if (!cvtOp)
return mlir::failure();
auto loadOp = dyn_cast_or_null<triton::LoadOp>(cvtOp.src().getDefiningOp());
if (!loadOp)
return mlir::failure();
auto dstTy = dstOp.getResult().getType().cast<RankedTensorType>();
auto srcTy = cvtOp.getOperand().getType().cast<RankedTensorType>();
if (dstTy != srcTy)
return mlir::failure();
// TODO: int tensor cores
auto _0f = rewriter.create<arith::ConstantFloatOp>(
op->getLoc(), APFloat(0.0f), dstTy.getElementType().cast<FloatType>());
auto _0 = rewriter.create<triton::SplatOp>(
op->getLoc(), dotOp.getResult().getType(), _0f);
auto newDot = rewriter.create<triton::DotOp>(
op->getLoc(), dotOp.getResult().getType(), dotOp.getOperand(0),
dotOp.getOperand(1), _0, dotOp.allowTF32());
auto newCvt = rewriter.create<triton::gpu::ConvertLayoutOp>(
op->getLoc(), dstTy, newDot.getResult());
auto newAdd = rewriter.replaceOpWithNewOp<arith::AddFOp>(
op, newCvt, cvtOp.getOperand());
return mlir::success(); return mlir::success();
} }
}; };
@@ -1477,6 +1528,7 @@ public:
patterns.add<MoveConvertOutOfIf>(context); patterns.add<MoveConvertOutOfIf>(context);
patterns.add<BlockedToMMA>(context, computeCapability); patterns.add<BlockedToMMA>(context, computeCapability);
patterns.add<ConvertTransConvert>(context); patterns.add<ConvertTransConvert>(context);
patterns.add<ConvertDotConvert>(context);
if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed()) { if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed()) {
signalPassFailure(); signalPassFailure();

2
python/slow.ttgir
View File

@@ -148,7 +148,7 @@ module attributes {"triton_gpu.num-warps" = 8 : i32} {
%136 = triton_gpu.convert_layout %60 : (tensor<128x64xf16, #shared0>) -> tensor<128x64xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #mma1}>> %136 = triton_gpu.convert_layout %60 : (tensor<128x64xf16, #shared0>) -> tensor<128x64xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #mma1}>>
%137 = tt.dot %135, %136, %134 {allowTF32 = true} : tensor<128x128xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #mma1}>> * tensor<128x64xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #mma1}>> -> tensor<128x64xf32, #mma1> %137 = tt.dot %135, %136, %134 {allowTF32 = true} : tensor<128x128xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #mma1}>> * tensor<128x64xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #mma1}>> -> tensor<128x64xf32, #mma1>
%138 = triton_gpu.convert_layout %137 : (tensor<128x64xf32, #mma1>) -> tensor<128x64xf32, #blocked2> %138 = triton_gpu.convert_layout %137 : (tensor<128x64xf32, #mma1>) -> tensor<128x64xf32, #blocked2>
tt.store %arg29, %138 : tensor<128x64xf32, #blocked2> tt.store %arg29, %133 : tensor<128x64xf32, #blocked2>
%139 = tt.addptr %arg29, %43 : tensor<128x64x!tt.ptr<f32>, #blocked2>, tensor<128x64xi32, #blocked2> %139 = tt.addptr %arg29, %43 : tensor<128x64x!tt.ptr<f32>, #blocked2>, tensor<128x64xi32, #blocked2>
%140 = tt.addptr %arg30, %42 : tensor<128x64x!tt.ptr<f16>, #blocked1>, tensor<128x64xi32, #blocked1> %140 = tt.addptr %arg30, %42 : tensor<128x64x!tt.ptr<f16>, #blocked1>, tensor<128x64xi32, #blocked1>
%141 = tt.addptr %arg31, %42 : tensor<128x64x!tt.ptr<f16>, #blocked1>, tensor<128x64xi32, #blocked1> %141 = tt.addptr %arg31, %42 : tensor<128x64x!tt.ptr<f16>, #blocked1>, tensor<128x64xi32, #blocked1>

3
python/tutorials/06-fused-attention.py
View File

@@ -191,6 +191,7 @@ def _bwd_kernel(
tl.store(dv_ptrs, dv) tl.store(dv_ptrs, dv)
tl.store(dk_ptrs, dk) tl.store(dk_ptrs, dk)
# _bwd_kernel = triton.compile("./slow.ttgir", num_warps=8) # _bwd_kernel = triton.compile("./slow.ttgir", num_warps=8)
# _bwd_kernel = triton.compile("./unoptimized.ttgir", num_warps=8) # _bwd_kernel = triton.compile("./unoptimized.ttgir", num_warps=8)
# _bwd_kernel = triton.compile("./bwd.ttgir", num_warps=8) # _bwd_kernel = triton.compile("./bwd.ttgir", num_warps=8)
@@ -260,7 +261,7 @@ class _attention(torch.autograd.Function):
BLOCK_M=ctx.BLOCK, D_HEAD=ctx.BLOCK_DMODEL, BLOCK_M=ctx.BLOCK, D_HEAD=ctx.BLOCK_DMODEL,
) )
# _bwd_kernel[(ctx.grid[1],1,1)]( # _bwd_kernel[(ctx.grid[1], 1, 1)](
# q.data_ptr(), k.data_ptr(), v.data_ptr(), ctx.sm_scale, # q.data_ptr(), k.data_ptr(), v.data_ptr(), ctx.sm_scale,
# o.data_ptr(), do_scaled.data_ptr(), # o.data_ptr(), do_scaled.data_ptr(),
# dq.data_ptr(), dk.data_ptr(), dv.data_ptr(), # dq.data_ptr(), dk.data_ptr(), dv.data_ptr(),