696f758435
Fix and improve swizzle loads for ordered indices
2021-06-08 14:19:27 +01:00
3eb42ecb55
Minor improvements to -use-separate-modules
2021-06-08 13:00:20 +01:00
28e9a4f79c
Replace js_wasm32 with freestanding_wasm32
2021-06-08 12:18:26 +01:00
8d8adac1b4
Fix lb_build_defer_stmt
2021-06-08 10:43:11 +01:00
963b1a12d7
Correct code for #simd in unions
2021-06-06 16:56:03 +01:00
cef16feb0b
Fix #861 - Add extra check on missing main
2021-06-06 12:46:59 +01:00
785c27daa7
Fix 128-bit integer to float cast by explicitly calling the procedure direct; Fix #781
2021-06-06 12:35:38 +01:00
795a5910cf
Add support for Addressing_OptionalOkPtr
...
Allowing for `i, ok := &x.(T);` (type assertions) and `v, ok := &m[k];` (map indexing)
2021-06-06 12:18:45 +01:00
4c21f9495d
Clean up lbAddr_Swizzle logic for load and store
2021-06-05 23:56:59 +01:00
f119fd1ee1
Use shufflevector when possible for lbAddr_Swizzle load
2021-06-05 22:50:23 +01:00
46ab822316
Improved lb_build_assign_stmt_array logic
2021-06-05 22:07:39 +01:00
a2f2041aa6
Fix lb_build_assign_stmt_array for lbAddr_Swizzle
2021-06-05 17:33:42 +01:00
599d18f26f
Experimental support for inline swizzling for array types of len <= 4 e.g. v.xyz, v.argb, v.xxx
2021-06-05 17:22:39 +01:00
8c943eb054
Make inline array arithmetic use load+extractvalue rather than getelementptr+load to give the optimizer a better hint for vectorization
2021-06-01 10:51:54 +01:00
446703ba75
Improves to array arithmetic on += etc assignment statements
2021-06-01 10:37:31 +01:00
2573da12fc
Fix #992
2021-05-31 20:58:28 +01:00
bbc9c6a93c
Fix #999
2021-05-31 20:56:07 +01:00
b8d6dd4eb7
Fix #1004
2021-05-31 20:38:10 +01:00
0f91ffe28f
Add intrinsics.{ptr_offset, ptr_sub}
2021-05-30 13:21:56 +01:00
4b46d691f8
Improve logic for intrinsics.{mem_copy, mem_copy_non_overlapping, mem_zero} to use the *.inline LLVM variants when possible
2021-05-30 12:52:44 +01:00
bb7bd94b0a
Fix comparison bug of enumerated arrays
2021-05-27 09:52:50 +01:00
4a886a1bc5
Disable copy elision on assignments for the time being
2021-05-25 15:43:34 +01:00
c21c754b6f
Minimize copying on getting the address of a call if required
2021-05-24 23:51:01 +01:00
0c46d06e63
Add intrinsics.mem_zero
2021-05-24 22:39:27 +01:00
44b6e7c45d
Move the mem zero into a separate procedure for reuse
2021-05-24 22:32:38 +01:00
284a2631fd
Refactoring of lbFunctionType retrieval
2021-05-24 22:16:22 +01:00
3f156bcb4b
Refactor backend code for assignments; Refactor some statements into separate procedures
2021-05-24 22:09:21 +01:00
d35a9e65b6
Heavily improve the copy elision logic in the backend
2021-05-24 20:57:44 +01:00
c440296ae8
Add @(link_section=<string>) for global variables
2021-05-24 15:41:22 +01:00
79f115d6a7
Handle #c_vararg correctly
2021-05-24 14:46:03 +01:00
b8f8d4c3a1
Modify ABI for the Odin calling conventions on SysV slightly
2021-05-22 11:33:08 +01:00
247f4f3293
Fix ..= logic in the backend
2021-05-21 10:51:19 +01:00
9c54ed5792
Add range-based error messages to -verbose-errors
...
Example:
Cannot convert '(1 + 2)' to 'untyped bool' from 'untyped integer'
x := (1 + 2) * true;
^~~~~~^
2021-05-19 14:15:57 +01:00
26ce40c188
Remove @(static) for global variables
2021-05-19 11:51:48 +01:00
b34e4a9fd1
More minor linkage changes
2021-05-19 11:46:27 +01:00
28561ef5f5
Minor change to internal linkage stuff
2021-05-19 11:26:05 +01:00
e82e4398b6
Add intrinsics.mem_copy and intrinsics.mem_copy_non_overlapping
2021-05-19 10:50:02 +01:00
e0225c3579
Add intrinsics.sqrt for floating-point values
2021-05-19 10:32:41 +01:00
6ef96d3300
Improve untyped to typed logic for aiding the backend
2021-05-16 14:44:02 +01:00
2e633f57a0
Add concrete type information for untyped values as procedure arguments
2021-05-16 13:00:16 +01:00
ce08e832f7
Allow ..= alongside .. as a "full range" operator; Update core:odin/parser etc
2021-05-16 12:34:35 +01:00
1cf6b6679d
Add custom basic dead instruction elimination pass
2021-05-15 21:10:06 +01:00
7b7081d607
Remove old dead code
2021-05-15 18:59:54 +01:00
5ae564cc8c
Add name to aggregate result pointer to procedures
2021-05-15 18:53:52 +01:00
f7b1290fe9
Change for i in x..y {} behaviour
...
Adds an extra check before incrementation to prevent the possibility of overflowing of `y` is at the limit maximum size of the integer
e.g. `for i in u8(0)..255 {}` (assuming `255` is not known at compile time)
2021-05-15 18:02:06 +01:00
b01c2e1017
Disallow slicing of constant values
2021-05-15 16:40:40 +01:00
63b54ce7c6
Add minor ignoring hint on type assertions to get better code generation with no optimizations enabled
2021-05-13 12:48:12 +01:00
2e5f57d8a1
Fix #741
2021-05-13 00:44:33 +01:00
d2fcbf0e1d
Fix #948
2021-05-12 15:00:19 +01:00
d353f97f91
Add byval with align, sret attributes for SysV
2021-05-11 13:30:27 +01:00
gingerBill