RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 19:23:23 +01:00
Code Issues Projects Releases Wiki Activity
369ae03621
llvm-mirror/test/CodeGen/X86/avx-intel-ocl.ll

173 lines
7.3 KiB
LLVM
Raw Normal View History

Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; RUN: llc < %s -mtriple=i686-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck -check-prefix=X32 %s
; RUN: llc < %s -mtriple=i386-pc-win32 -mcpu=corei7-avx -mattr=+avx | FileCheck -check-prefix=X32 %s
The test avx-intel-ocl.ll failed. I can't reproduce on any of my machines. I added -mcpu flag, may be it will fix the problem llvm-svn: 166669
2012-10-25 10:38:42 +02:00
; RUN: llc < %s -mtriple=x86_64-win32 -mcpu=corei7-avx -mattr=+avx | FileCheck -check-prefix=WIN64 %s
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck -check-prefix=X64 %s
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
declare <16 x float> @func_float16_ptr(<16 x float>, <16 x float> *)
declare <16 x float> @func_float16(<16 x float>, <16 x float>)
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
declare i32 @func_int(i32, i32)
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; WIN64-LABEL: testf16_inp
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
; WIN64: vaddps {{.*}}, {{%ymm[0-1]}}
; WIN64: vaddps {{.*}}, {{%ymm[0-1]}}
; WIN64: leaq {{.*}}(%rsp), %rcx
; WIN64: call
; WIN64: ret
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; X32-LABEL: testf16_inp
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; X32: movl %eax, (%esp)
; X32: vaddps {{.*}}, {{%ymm[0-1]}}
; X32: vaddps {{.*}}, {{%ymm[0-1]}}
; X32: call
; X32: ret
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; X64-LABEL: testf16_inp
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; X64: vaddps {{.*}}, {{%ymm[0-1]}}
; X64: vaddps {{.*}}, {{%ymm[0-1]}}
; X64: leaq {{.*}}(%rsp), %rdi
; X64: call
; X64: ret
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
;test calling conventions - input parameters
define <16 x float> @testf16_inp(<16 x float> %a, <16 x float> %b) nounwind {
%y = alloca <16 x float>, align 16
%x = fadd <16 x float> %a, %b
Enable MI Sched for x86. This changes the SelectionDAG scheduling preference to source order. Soon, the SelectionDAG scheduler can be bypassed saving a nice chunk of compile time. Performance differences that result from this change are often a consequence of register coalescing. The register coalescer is far from perfect. Bugs can be filed for deficiencies. On x86 SandyBridge/Haswell, the source order schedule is often preserved, particularly for small blocks. Register pressure is generally improved over the SD scheduler's ILP mode. However, we are still able to handle large blocks that require latency hiding, unlike the SD scheduler's BURR mode. MI scheduler also attempts to discover the critical path in single-block loops and adjust heuristics accordingly. The MI scheduler relies on the new machine model. This is currently unimplemented for AVX, so we may not be generating the best code yet. Unit tests are updated so they don't depend on SD scheduling heuristics. llvm-svn: 192750
2013-10-16 01:33:07 +02:00
%1 = call intel_ocl_bicc <16 x float> @func_float16_ptr(<16 x float> %x, <16 x float>* %y)
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
%2 = load <16 x float>* %y, align 16
%3 = fadd <16 x float> %2, %1
ret <16 x float> %3
}
;test calling conventions - preserved registers
; preserved ymm6-ymm15
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; WIN64-LABEL: testf16_regs
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
; WIN64: call
Enable MI Sched for x86. This changes the SelectionDAG scheduling preference to source order. Soon, the SelectionDAG scheduler can be bypassed saving a nice chunk of compile time. Performance differences that result from this change are often a consequence of register coalescing. The register coalescer is far from perfect. Bugs can be filed for deficiencies. On x86 SandyBridge/Haswell, the source order schedule is often preserved, particularly for small blocks. Register pressure is generally improved over the SD scheduler's ILP mode. However, we are still able to handle large blocks that require latency hiding, unlike the SD scheduler's BURR mode. MI scheduler also attempts to discover the critical path in single-block loops and adjust heuristics accordingly. The MI scheduler relies on the new machine model. This is currently unimplemented for AVX, so we may not be generating the best code yet. Unit tests are updated so they don't depend on SD scheduling heuristics. llvm-svn: 192750
2013-10-16 01:33:07 +02:00
; WIN64: vaddps {{%ymm[6-7]}}, {{%ymm[0-1]}}, {{%ymm[0-1]}}
; WIN64: vaddps {{%ymm[6-7]}}, {{%ymm[0-1]}}, {{%ymm[0-1]}}
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
; WIN64: ret
; preserved ymm8-ymm15
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; X64-LABEL: testf16_regs
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; X64: call
Enable MI Sched for x86. This changes the SelectionDAG scheduling preference to source order. Soon, the SelectionDAG scheduler can be bypassed saving a nice chunk of compile time. Performance differences that result from this change are often a consequence of register coalescing. The register coalescer is far from perfect. Bugs can be filed for deficiencies. On x86 SandyBridge/Haswell, the source order schedule is often preserved, particularly for small blocks. Register pressure is generally improved over the SD scheduler's ILP mode. However, we are still able to handle large blocks that require latency hiding, unlike the SD scheduler's BURR mode. MI scheduler also attempts to discover the critical path in single-block loops and adjust heuristics accordingly. The MI scheduler relies on the new machine model. This is currently unimplemented for AVX, so we may not be generating the best code yet. Unit tests are updated so they don't depend on SD scheduling heuristics. llvm-svn: 192750
2013-10-16 01:33:07 +02:00
; X64: vaddps {{%ymm[8-9]}}, {{%ymm[0-1]}}, {{%ymm[0-1]}}
; X64: vaddps {{%ymm[8-9]}}, {{%ymm[0-1]}}, {{%ymm[0-1]}}
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; X64: ret
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
define <16 x float> @testf16_regs(<16 x float> %a, <16 x float> %b) nounwind {
%y = alloca <16 x float>, align 16
%x = fadd <16 x float> %a, %b
Enable MI Sched for x86. This changes the SelectionDAG scheduling preference to source order. Soon, the SelectionDAG scheduler can be bypassed saving a nice chunk of compile time. Performance differences that result from this change are often a consequence of register coalescing. The register coalescer is far from perfect. Bugs can be filed for deficiencies. On x86 SandyBridge/Haswell, the source order schedule is often preserved, particularly for small blocks. Register pressure is generally improved over the SD scheduler's ILP mode. However, we are still able to handle large blocks that require latency hiding, unlike the SD scheduler's BURR mode. MI scheduler also attempts to discover the critical path in single-block loops and adjust heuristics accordingly. The MI scheduler relies on the new machine model. This is currently unimplemented for AVX, so we may not be generating the best code yet. Unit tests are updated so they don't depend on SD scheduling heuristics. llvm-svn: 192750
2013-10-16 01:33:07 +02:00
%1 = call intel_ocl_bicc <16 x float> @func_float16_ptr(<16 x float> %x, <16 x float>* %y)
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
%2 = load <16 x float>* %y, align 16
%3 = fadd <16 x float> %1, %b
%4 = fadd <16 x float> %2, %3
ret <16 x float> %4
}
; test calling conventions - prolog and epilog
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; WIN64-LABEL: test_prolog_epilog
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
; WIN64: vmovaps {{%ymm([6-9]|1[0-5])}}, {{.*(%rbp).*}} # 32-byte Spill
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
; WIN64: call
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; WIN64: vmovaps {{.*(%rbp).*}}, {{%ymm([6-9]|1[0-5])}} # 32-byte Reload
; X64-LABEL: test_prolog_epilog
[X86][AVX] 256-bit vector stack unaligned load/stores identification Under many circumstances the stack is not 32-byte aligned, resulting in the use of the vmovups/vmovupd/vmovdqu instructions when inserting ymm reloads/spills. This minor patch adds these instructions to the isFrameLoadOpcode/isFrameStoreOpcode helpers so that they can be correctly identified and not be treated as folded reloads/spills. This has also been noticed by http://llvm.org/bugs/show_bug.cgi?id=18846 where it was causing redundant spills - I've added a reduced test case at test/CodeGen/X86/pr18846.ll Differential Revision: http://reviews.llvm.org/D6252 llvm-svn: 222281
2014-11-19 00:38:19 +01:00
; X64: vmovups {{%ymm([8-9]|1[0-5])}}, {{.*}}(%rsp) ## 32-byte Spill
; X64: vmovups {{%ymm([8-9]|1[0-5])}}, {{.*}}(%rsp) ## 32-byte Spill
; X64: vmovups {{%ymm([8-9]|1[0-5])}}, {{.*}}(%rsp) ## 32-byte Spill
; X64: vmovups {{%ymm([8-9]|1[0-5])}}, {{.*}}(%rsp) ## 32-byte Spill
; X64: vmovups {{%ymm([8-9]|1[0-5])}}, {{.*}}(%rsp) ## 32-byte Spill
; X64: vmovups {{%ymm([8-9]|1[0-5])}}, {{.*}}(%rsp) ## 32-byte Spill
; X64: vmovups {{%ymm([8-9]|1[0-5])}}, {{.*}}(%rsp) ## 32-byte Spill
; X64: vmovups {{%ymm([8-9]|1[0-5])}}, {{.*}}(%rsp) ## 32-byte Spill
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; X64: call
[X86][AVX] 256-bit vector stack unaligned load/stores identification Under many circumstances the stack is not 32-byte aligned, resulting in the use of the vmovups/vmovupd/vmovdqu instructions when inserting ymm reloads/spills. This minor patch adds these instructions to the isFrameLoadOpcode/isFrameStoreOpcode helpers so that they can be correctly identified and not be treated as folded reloads/spills. This has also been noticed by http://llvm.org/bugs/show_bug.cgi?id=18846 where it was causing redundant spills - I've added a reduced test case at test/CodeGen/X86/pr18846.ll Differential Revision: http://reviews.llvm.org/D6252 llvm-svn: 222281
2014-11-19 00:38:19 +01:00
; X64: vmovups {{.*}}(%rsp), {{%ymm([8-9]|1[0-5])}} ## 32-byte Reload
; X64: vmovups {{.*}}(%rsp), {{%ymm([8-9]|1[0-5])}} ## 32-byte Reload
; X64: vmovups {{.*}}(%rsp), {{%ymm([8-9]|1[0-5])}} ## 32-byte Reload
; X64: vmovups {{.*}}(%rsp), {{%ymm([8-9]|1[0-5])}} ## 32-byte Reload
; X64: vmovups {{.*}}(%rsp), {{%ymm([8-9]|1[0-5])}} ## 32-byte Reload
; X64: vmovups {{.*}}(%rsp), {{%ymm([8-9]|1[0-5])}} ## 32-byte Reload
; X64: vmovups {{.*}}(%rsp), {{%ymm([8-9]|1[0-5])}} ## 32-byte Reload
; X64: vmovups {{.*}}(%rsp), {{%ymm([8-9]|1[0-5])}} ## 32-byte Reload
Special calling conventions for Intel OpenCL built-in library. llvm-svn: 166566
2012-10-24 16:46:16 +02:00
define intel_ocl_bicc <16 x float> @test_prolog_epilog(<16 x float> %a, <16 x float> %b) nounwind {
%c = call <16 x float> @func_float16(<16 x float> %a, <16 x float> %b)
ret <16 x float> %c
The test avx-intel-ocl.ll failed. I can't reproduce on any of my machines. I added -mcpu flag, may be it will fix the problem llvm-svn: 166669
2012-10-25 10:38:42 +02:00
}
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; test functions with integer parameters
; pass parameters on stack for 32-bit platform
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; X32-LABEL: test_int
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; X32: movl {{.*}}, 4(%esp)
; X32: movl {{.*}}, (%esp)
; X32: call
; X32: addl {{.*}}, %eax
; pass parameters in registers for 64-bit platform
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; X64-LABEL: test_int
Intel OCL built-ins calling conventions now support MacOS 32-bit. llvm-svn: 168359
2012-11-20 10:37:57 +01:00
; X64: leal {{.*}}, %edi
; X64: movl {{.*}}, %esi
; X64: call
; X64: addl {{.*}}, %eax
define i32 @test_int(i32 %a, i32 %b) nounwind {
%c1 = add i32 %a, %b
%c2 = call intel_ocl_bicc i32 @func_int(i32 %c1, i32 %a)
%c = add i32 %c2, %b
ret i32 %c
}
Prevent insertion of "vzeroupper" before call that preserves YMM registers, since a caller uses preserved registers across the call. llvm-svn: 175043
2013-02-13 09:02:04 +01:00
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; WIN64-LABEL: test_float4
Prevent insertion of "vzeroupper" before call that preserves YMM registers, since a caller uses preserved registers across the call. llvm-svn: 175043
2013-02-13 09:02:04 +01:00
; WIN64-NOT: vzeroupper
; WIN64: call
; WIN64-NOT: vzeroupper
; WIN64: call
; WIN64: ret
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; X64-LABEL: test_float4
Prevent insertion of "vzeroupper" before call that preserves YMM registers, since a caller uses preserved registers across the call. llvm-svn: 175043
2013-02-13 09:02:04 +01:00
; X64-NOT: vzeroupper
; X64: call
; X64-NOT: vzeroupper
; X64: call
; X64: ret
Re-apply r211399, "Generate native unwind info on Win64" with a fix to ignore SEH pseudo ops in X86 JIT emitter. -- This patch enables LLVM to emit Win64-native unwind info rather than DWARF CFI. It handles all corner cases (I hope), including stack realignment. Because the unwind info is not flexible enough to describe stack frames with a gap of unknown size in the middle, such as the one caused by stack realignment, I modified register spilling code to place all spills into the fixed frame slots, so that they can be accessed relative to the frame pointer. Patch by Vadim Chugunov! Reviewed By: rnk Differential Revision: http://reviews.llvm.org/D4081 llvm-svn: 211691
2014-06-25 14:41:52 +02:00
; X32-LABEL: test_float4
Prevent insertion of "vzeroupper" before call that preserves YMM registers, since a caller uses preserved registers across the call. llvm-svn: 175043
2013-02-13 09:02:04 +01:00
; X32: vzeroupper
; X32: call
; X32: vzeroupper
; X32: call
; X32: ret
declare <4 x float> @func_float4(<4 x float>, <4 x float>, <4 x float>)
define <8 x float> @test_float4(<8 x float> %a, <8 x float> %b, <8 x float> %c) nounwind readnone {
entry:
%0 = shufflevector <8 x float> %a, <8 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%1 = shufflevector <8 x float> %b, <8 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%2 = shufflevector <8 x float> %c, <8 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%call.i = tail call intel_ocl_bicc <4 x float> @func_float4(<4 x float> %0, <4 x float> %1, <4 x float> %2) nounwind
%3 = shufflevector <4 x float> %call.i, <4 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
%4 = shufflevector <8 x float> %a, <8 x float> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
%5 = shufflevector <8 x float> %b, <8 x float> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
%6 = shufflevector <8 x float> %c, <8 x float> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
%call.i2 = tail call intel_ocl_bicc <4 x float> @func_float4(<4 x float> %4, <4 x float> %5, <4 x float> %6) nounwind
%7 = shufflevector <4 x float> %call.i2, <4 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
%8 = shufflevector <8 x float> %3, <8 x float> %7, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
ret <8 x float> %8
}
Reference in New Issue Copy Permalink