make SkJumper stages normal Skia code

Enough clients are using Clang now that we can say, use Clang to build
if you want these software pipeline stages to go fast.

This lets us drop the offline build aspect of SkJumper stages, instead
building as part of Skia using the SkOpts framework.

I think everything should work, except I've (temporarily) removed
AVX-512 support.  I will put this back in a follow up.

I have had to drop Windows down to __vectorcall and our narrower
stage calling convention that keeps the d-registers on the stack.
I tried forcing sysv_abi, but that crashed Clang.  :/

Added a TODO to up the same narrower stage calling convention
for lowp stages... we just *don't* today, for no good reason.

Change-Id: Iaaa792ffe4deab3508d2dc5d0008c163c24b3383
Reviewed-on: https://skia-review.googlesource.com/110641
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Herb Derby <herb@google.com>
Reviewed-by: Florin Malita <fmalita@chromium.org>

1 files changed, 6 insertions, 29 deletions

diff --git a/src/jumper/SkJumper.h b/src/jumper/SkJumper.h
index e7bf9b2403..d4c4698048 100644
--- a/src/jumper/SkJumper.h
+++ b/src/jumper/SkJumper.h
@@ -11,34 +11,11 @@
 #include <stddef.h>
 #include <stdint.h>
 
-// This file contains definitions shared by SkJumper.cpp (compiled normally as part of Skia)
-// and SkJumper_stages.cpp (compiled into Skia _and_ offline into SkJumper_generated.h).
-// Keep it simple!
-
-// Externally facing functions (start_pipeline) are called a little specially on Windows.
-#if defined(JUMPER_IS_OFFLINE) && defined(WIN) && defined(__x86_64__)
-    #define MAYBE_MSABI __attribute__((ms_abi))                   // Use MS' ABI, not System V.
-#elif defined(JUMPER_IS_OFFLINE) && defined(WIN) && defined(__i386__)
-    #define MAYBE_MSABI __attribute__((force_align_arg_pointer))  // Re-align stack 4 -> 16 bytes.
-#else
-    #define MAYBE_MSABI
-#endif
-
-// Any custom ABI to use for all non-externally-facing stage functions.
-#if defined(__ARM_NEON) && defined(__arm__)
-    // This lets us pass vectors more efficiently on 32-bit ARM.
-    #define ABI __attribute__((pcs("aapcs-vfp")))
-#else
-    #define ABI
-#endif
-
-// On ARM we expect that you're using Clang if you want SkJumper to be fast.
-// If you are, the baseline float stages will use NEON, and lowp stages will
-// also be available. (If somehow you're building for ARM not using Clang,
-// you'll get scalar baseline stages and no lowp support.)
-#if defined(__clang__) && defined(__ARM_NEON)
-    #define JUMPER_HAS_NEON_LOWP
-#endif
+// This file contains definitions shared by SkJumper.cpp/SkJumper_stages.cpp
+// and the rest of Skia.  It is important to keep the interface to SkJumper
+// limited and simple to avoid serious ODR violation pitfalls, especially when
+// using Microsoft's <math.h> and similar headers with inline-but-not-static
+// function definitions.
 
 static const int SkJumper_kMaxStride = 16;
 
@@ -76,7 +53,7 @@ struct SkJumper_DecalTileCtx {
 };
 
 struct SkJumper_CallbackCtx {
-    MAYBE_MSABI void (*fn)(SkJumper_CallbackCtx* self, int active_pixels/*<= SkJumper_kMaxStride*/);
+    void (*fn)(SkJumper_CallbackCtx* self, int active_pixels/*<= SkJumper_kMaxStride*/);
 
     // When called, fn() will have our active pixels available in rgba.
     // When fn() returns, the pipeline will read back those active pixels from read_from.