diff options
| author | Abhiroop Sarkar <asiamgenius@gmail.com> | 2018-09-27 15:28:46 -0400 |
|---|---|---|
| committer | Marge Bot <ben+marge-bot@smart-cactus.org> | 2019-07-03 09:33:39 -0400 |
| commit | acd795583625401c5554f8e04ec7efca18814011 (patch) | |
| tree | 545e529eed21e78592ff326d4ebf9804095ad2cb /compiler/cmm/CmmMachOp.hs | |
| parent | df3e5b744db29c085f5bc05f8b609197bcbf9b0c (diff) | |
| download | haskell-acd795583625401c5554f8e04ec7efca18814011.tar.gz | |
Add support for SIMD operations in the NCG
This adds support for constructing vector types from Float#, Double# etc
and performing arithmetic operations on them
Cleaned-Up-By: Ben Gamari <ben@well-typed.com>
Diffstat (limited to 'compiler/cmm/CmmMachOp.hs')
| -rw-r--r-- | compiler/cmm/CmmMachOp.hs | 25 |
1 files changed, 16 insertions, 9 deletions
diff --git a/compiler/cmm/CmmMachOp.hs b/compiler/cmm/CmmMachOp.hs index 9740d21bef..38d9edb480 100644 --- a/compiler/cmm/CmmMachOp.hs +++ b/compiler/cmm/CmmMachOp.hs @@ -136,8 +136,9 @@ data MachOp | MO_VU_Rem Length Width -- Floting point vector element insertion and extraction operations - | MO_VF_Insert Length Width -- Insert scalar into vector - | MO_VF_Extract Length Width -- Extract scalar from vector + | MO_VF_Broadcast Length Width -- Broadcast a scalar into a vector + | MO_VF_Insert Length Width -- Insert scalar into vector + | MO_VF_Extract Length Width -- Extract scalar from vector -- Floating point vector operations | MO_VF_Add Length Width @@ -430,6 +431,7 @@ machOpResultType dflags mop tys = MO_VU_Quot l w -> cmmVec l (cmmBits w) MO_VU_Rem l w -> cmmVec l (cmmBits w) + MO_VF_Broadcast l w -> cmmVec l (cmmFloat w) MO_VF_Insert l w -> cmmVec l (cmmFloat w) MO_VF_Extract _ w -> cmmFloat w @@ -522,16 +524,21 @@ machOpArgReps dflags op = MO_VU_Quot _ r -> [r,r] MO_VU_Rem _ r -> [r,r] - MO_VF_Insert l r -> [typeWidth (vec l (cmmFloat r)),r,wordWidth dflags] - MO_VF_Extract l r -> [typeWidth (vec l (cmmFloat r)),wordWidth dflags] + -- offset is always W32 as mentioned in StgCmmPrim.hs + MO_VF_Broadcast l r -> [vecwidth l r, r] + MO_VF_Insert l r -> [vecwidth l r, r, W32] + MO_VF_Extract l r -> [vecwidth l r, W32] - MO_VF_Add _ r -> [r,r] - MO_VF_Sub _ r -> [r,r] - MO_VF_Mul _ r -> [r,r] - MO_VF_Quot _ r -> [r,r] - MO_VF_Neg _ r -> [r] + -- NOTE: The below is owing to the fact that floats use the SSE registers + MO_VF_Add l w -> [vecwidth l w, vecwidth l w] + MO_VF_Sub l w -> [vecwidth l w, vecwidth l w] + MO_VF_Mul l w -> [vecwidth l w, vecwidth l w] + MO_VF_Quot l w -> [vecwidth l w, vecwidth l w] + MO_VF_Neg l w -> [vecwidth l w] MO_AlignmentCheck _ r -> [r] + where + vecwidth l w = widthFromBytes (l*widthInBytes w) ----------------------------------------------------------------------------- -- CallishMachOp |