update simd macro to just call procs always

src/pixie/simd/avx2.nim

@@ -415,7 +415,7 @@ proc minifyBy2Avx2*(image: Image, power = 1): Image {.simd.} =
     # Set src as this result for if we do another power
     src = result
 
-template applyCoverage*(rgbxVec: M256i, coverage: M128i): M256i =
+template applyCoverage(rgbxVec: M256i, coverage: M128i): M256i =
   ## Unpack the first 8 coverage bytes.
   let
     unpacked0 = mm_shuffle_epi8(coverage, coverageShuffle)

src/pixie/simd/internal.nim

@@ -63,46 +63,58 @@ macro simd*(procedure: untyped) =
 macro hasSimd*(procedure: untyped) =
   let
     name = procedure.procName()
-    originalBody = procedure[6]
     nameNeon = name & "Neon"
     nameSse2 = name & "Sse2"
     nameAvx = name & "Avx"
     nameAvx2 = name & "Avx2"
+    callNeon = callAndReturn(ident(nameNeon), procedure)
+    callSse2 = callAndReturn(ident(nameSse2), procedure)
     callAvx = callAndReturn(ident(nameAvx), procedure)
     callAvx2 = callAndReturn(ident(nameAvx2), procedure)
 
   var
     foundSimd: bool
-    body = newStmtList()
+
+  if procedure[6].kind != nnkStmtList:
+    error("hasSimd proc body must start with nnkStmtList")
+
+  var insertIdx = 0
+  if procedure[6][0].kind == nnkCommentStmt:
+    insertIdx = 1
 
   when defined(amd64) and not defined(pixieNoAvx):
     if nameAvx2 & procSignature(procedure) in simdProcs:
       foundSimd = true
-      body.add quote do:
+      procedure[6].insert(insertIdx, quote do:
         if cpuHasAvx2:
           `callAvx2`
-
+      )
+      inc insertIdx
     if nameAvx & procSignature(procedure) in simdProcs:
       foundSimd = true
-      body.add quote do:
-        if cpuHasAvx2:
+      procedure[6].insert(insertIdx, quote do:
+        if cpuHasAvx:
           `callAvx`
-
-  if nameSse2 & procSignature(procedure) in simdProcs:
-    foundSimd = true
-    let bodySse2 = simdProcs[nameSse2 & procSignature(procedure)][6]
-    body.add quote do:
-      `bodySse2`
-  elif nameNeon & procSignature(procedure) in simdProcs:
-    foundSimd = true
-    let bodyNeon = simdProcs[nameNeon & procSignature(procedure)][6]
-    body.add quote do:
-      `bodyNeon`
-  else:
-    body.add quote do:
-      `originalBody`
-
-  procedure[6] = body
+      )
+      inc insertIdx
+  when defined(amd64):
+    if nameSse2 & procSignature(procedure) in simdProcs:
+      foundSimd = true
+      procedure[6].insert(insertIdx, quote do:
+        `callSse2`
+      )
+      inc insertIdx
+      while procedure[6].len > insertIdx:
+        procedure[6].del(insertIdx)
+  elif defined(arm64):
+    if nameNeon & procSignature(procedure) in simdProcs:
+      foundSimd = true
+      procedure[6].insert(insertIdx, quote do:
+        `callNeon`
+      )
+      inc insertIdx
+      while procedure[6].len > insertIdx:
+        procedure[6].del(insertIdx)
 
   when not defined(pixieNoSimd):
     if not foundSimd:

src/pixie/simd/neon.nim

@@ -3,17 +3,17 @@ import chroma, internal, nimsimd/neon, pixie/blends, pixie/common, vmath
 when defined(release):
   {.push checks: off.}
 
-template multiplyDiv255*(c, a: uint8x8): uint8x8 =
+template multiplyDiv255(c, a: uint8x8): uint8x8 =
   let ca = vmull_u8(c, a)
   vraddhn_u16(ca, vrshrq_n_u16(ca, 8))
 
-template multiplyDiv255*(c, a: uint8x16): uint8x16 =
+template multiplyDiv255(c, a: uint8x16): uint8x16 =
   vcombine_u8(
     multiplyDiv255(vget_low_u8(c), vget_low_u8(a)),
     multiplyDiv255(vget_high_u8(c), vget_high_u8(a))
   )
 
-template blendNormalSimd*(backdrop, source: uint8x16x4): uint8x16x4 =
+template blendNormalSimd(backdrop, source: uint8x16x4): uint8x16x4 =
   let multiplier = vsubq_u8(vec255, source.val[3])
 
   var blended: uint8x16x4

src/pixie/simd/sse2.nim

@@ -10,7 +10,7 @@ proc applyOpacity*(color: M128, opacity: float32): ColorRGBX {.inline.} =
   finalColor = mm_packus_epi16(finalColor, mm_setzero_si128())
   cast[ColorRGBX](mm_cvtsi128_si32(finalColor))
 
-template blendNormalSimd*(backdrop, source: M128i): M128i =
+template blendNormalSimd(backdrop, source: M128i): M128i =
   var
     sourceAlpha = mm_and_si128(source, alphaMask)
     backdropEven = mm_slli_epi16(backdrop, 8)
@@ -30,7 +30,7 @@ template blendNormalSimd*(backdrop, source: M128i): M128i =
     mm_or_si128(backdropEven, mm_slli_epi16(backdropOdd, 8))
   )
 
-template blendMaskSimd*(backdrop, source: M128i): M128i =
+template blendMaskSimd(backdrop, source: M128i): M128i =
   var
     sourceAlpha = mm_and_si128(source, alphaMask)
     backdropEven = mm_slli_epi16(backdrop, 8)
@@ -507,7 +507,7 @@ proc magnifyBy2Sse2*(image: Image, power = 1): Image {.simd.} =
         result.width * 4
       )
 
-template applyCoverage*(rgbxVec, coverage: M128i): M128i =
+template applyCoverage(rgbxVec, coverage: M128i): M128i =
   ## Unpack the first 4 coverage bytes.
   var unpacked = mm_unpacklo_epi8(mm_setzero_si128(), coverage)
   unpacked = mm_unpacklo_epi8(mm_setzero_si128(), unpacked)