T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add a DynRex recipe type for x86, decreasing the number of recipes (#1298)

Browse Source 
This patch adds a third mode for templates: REX inference is requestable
at template instantiation time. This reduces the number of recipes
by removing rex()/nonrex() redundancy for many instructions.
This commit is contained in:
Sean Stangl
2019-12-19 15:49:34 -07:00
committed by GitHub
parent b486289ab8
commit cf9e762f16
13 changed files with 875 additions and 514 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
cranelift/filetests/filetests/isa/x86/legalize-br-icmp.clif
View File

@@ -15,7 +15,7 @@ ebb1:
; sameln: function %br_icmp(i64 [%rdi]) fast {
; nextln: ebb0(v0: i64):
; nextln: [RexOp1pu_id#b8] v1 = iconst.i64 0
; nextln: [RexOp1icscc#8039] v2 = icmp eq v0, v1
; nextln: [DynRexOp1icscc#8039] v2 = icmp eq v0, v1
; nextln: [RexOp1t8jccb#75] brnz v2, ebb1
; nextln: [Op1jmpb#eb] jump ebb1
; nextln:
@@ -37,7 +37,7 @@ ebb1(v2: i64):
; sameln: function %br_icmp_ebb_args(i64 [%rdi]) fast {
; nextln: ebb0(v0: i64):
; nextln: [RexOp1pu_id#b8] v1 = iconst.i64 0
; nextln: [RexOp1icscc#8039] v3 = icmp eq v0, v1
; nextln: [DynRexOp1icscc#8039] v3 = icmp eq v0, v1
; nextln: [RexOp1t8jccb#75] brnz v3, ebb1(v0)
; nextln: [Op1jmpb#eb] jump ebb1(v0)
; nextln:

122
cranelift/filetests/filetests/isa/x86/relax_branch.clif
View File

@@ -21,95 +21,95 @@ function u0:2691(i32 [%rdi], i32 [%rsi], i64 vmctx [%r14]) -> i64 uext [%rax] ba
@0005 [-] fallthrough ebb3(v0, v1)
ebb3(v8: i32 [%rdi], v19: i32 [%rsi]):
@0005 [RexOp1ldDisp8#808b,%rax] v7 = load.i64 v2+48
@0005 [RexOp1rcmp_ib#f083,%rflags] v91 = ifcmp_imm v7, 0
@0005 [trapif#00] trapif ne v91, interrupt
[Op1umr#89,%rax] v105 = copy v8
@000b [Op1r_ib#83,%rax] v10 = iadd_imm v105, 1
v80 -> v10
@0010 [Op1umr#89,%rcx] v92 = uextend.i64 v8
@0010 [RexOp1ld#808b,%rdx] v93 = load.i64 notrap aligned readonly v2
v95 -> v93
@0010 [Op2ldWithIndex#4be,%rcx] v12 = sload8_complex.i32 v93+v92
[Op1umr#89,%rbx] v106 = copy v12
@0017 [Op1r_ib#40c1,%rbx] v14 = ishl_imm v106, 24
@001a [Op1r_ib#70c1,%rbx] v16 = sshr_imm v14, 24
[Op1umr#89,%rdi] v107 = copy v16
@001f [Op1r_ib#83,%rdi] v18 = iadd_imm v107, 32
[RexOp1umr#89,%r8] v108 = copy v19
@0026 [RexOp1r_ib#83,%r8] v21 = iadd_imm v108, 1
v82 -> v21
@002b [Op1umr#89,%rsi] v94 = uextend.i64 v19
@002b [Op2ldWithIndex#4be,%rdx] v23 = sload8_complex.i32 v93+v94
v55 -> v23
[Op1umr#89,%rsi] v109 = copy v23
@0032 [Op1r_ib#40c1,%rsi] v25 = ishl_imm v109, 24
@0035 [Op1r_ib#70c1,%rsi] v27 = sshr_imm v25, 24
v69 -> v27
[RexOp1umr#89,%r9] v110 = copy v27
@003a [RexOp1r_ib#83,%r9] v29 = iadd_imm v110, 32
v68 -> v29
@0042 [Op1r_ib#83,%rcx] v31 = iadd_imm v12, -65
@0045 [Op1r_ib#40c1,%rcx] v33 = ishl_imm v31, 24
@0048 [Op1r_ib#70c1,%rcx] v35 = sshr_imm v33, 24
@004c [Op1r_id#4081,%rcx] v37 = band_imm v35, 255
[Op1rcmp_ib#7083,%rflags] v97 = ifcmp_imm v37, 26
@0050 [Op1brib#70] brif sge v97, ebb6
@0050 [-] fallthrough ebb10
@0005 [RexOp1ldDisp8#808b,%rax] v7 = load.i64 v2+48
@0005 [DynRexOp1rcmp_ib#f083,%rflags] v91 = ifcmp_imm v7, 0
@0005 [trapif#00] trapif ne v91, interrupt
[DynRexOp1umr#89,%rax] v105 = copy v8
@000b [DynRexOp1r_ib#83,%rax] v10 = iadd_imm v105, 1
v80 -> v10
@0010 [Op1umr#89,%rcx] v92 = uextend.i64 v8
@0010 [RexOp1ld#808b,%rdx] v93 = load.i64 notrap aligned readonly v2
v95 -> v93
@0010 [Op2ldWithIndex#4be,%rcx] v12 = sload8_complex.i32 v93+v92
[DynRexOp1umr#89,%rbx] v106 = copy v12
@0017 [DynRexOp1r_ib#40c1,%rbx] v14 = ishl_imm v106, 24
@001a [DynRexOp1r_ib#70c1,%rbx] v16 = sshr_imm v14, 24
[DynRexOp1umr#89,%rdi] v107 = copy v16
@001f [DynRexOp1r_ib#83,%rdi] v18 = iadd_imm v107, 32
[DynRexOp1umr#89,%r8] v108 = copy v19
@0026 [DynRexOp1r_ib#83,%r8] v21 = iadd_imm v108, 1
v82 -> v21
@002b [Op1umr#89,%rsi] v94 = uextend.i64 v19
@002b [Op2ldWithIndex#4be,%rdx] v23 = sload8_complex.i32 v93+v94
v55 -> v23
[DynRexOp1umr#89,%rsi] v109 = copy v23
@0032 [DynRexOp1r_ib#40c1,%rsi] v25 = ishl_imm v109, 24
@0035 [DynRexOp1r_ib#70c1,%rsi] v27 = sshr_imm v25, 24
v69 -> v27
[DynRexOp1umr#89,%r9] v110 = copy v27
@003a [DynRexOp1r_ib#83,%r9] v29 = iadd_imm v110, 32
v68 -> v29
@0042 [DynRexOp1r_ib#83,%rcx] v31 = iadd_imm v12, -65
@0045 [DynRexOp1r_ib#40c1,%rcx] v33 = ishl_imm v31, 24
@0048 [DynRexOp1r_ib#70c1,%rcx] v35 = sshr_imm v33, 24
@004c [DynRexOp1r_id#4081,%rcx] v37 = band_imm v35, 255
[DynRexOp1rcmp_ib#7083,%rflags] v97 = ifcmp_imm v37, 26
@0050 [Op1brib#70] brif sge v97, ebb6
@0050 [-] fallthrough ebb10
ebb10:
[Op1umr#89,%rcx] v101 = copy v18
[DynRexOp1umr#89,%rcx] v101 = copy v18
@0054 [Op1jmpb#eb] jump ebb5(v18, v101)
ebb6:
[Op1umr#89,%rcx] v102 = copy.i32 v16
[DynRexOp1umr#89,%rcx] v102 = copy.i32 v16
@0059 [RexOp1rmov#89] regmove v102, %rcx -> %rdi
@0059 [RexOp1rmov#89] regmove.i32 v16, %rbx -> %rcx
@0059 [-] fallthrough ebb5(v102, v16)
ebb5(v41: i32 [%rdi], v84: i32 [%rcx]):
v83 -> v84
@005d [Op1r_id#4081,%rdi] v43 = band_imm v41, 255
@0062 [Op1r_ib#40c1,%rdi] v45 = ishl_imm v43, 24
@005d [DynRexOp1r_id#4081,%rdi] v43 = band_imm v41, 255
@0062 [DynRexOp1r_ib#40c1,%rdi] v45 = ishl_imm v43, 24
v52 -> v45
@0065 [RexOp1rmov#89] regmove v45, %rdi -> %rbx
@0065 [Op1r_ib#70c1,%rbx] v47 = sshr_imm v45, 24
@0065 [DynRexOp1r_ib#70c1,%rbx] v47 = sshr_imm v45, 24
v54 -> v47
@0068 [RexOp1rmov#89] regmove v47, %rbx -> %rdi
@0068 [Op1icscc_ib#7083,%rbx] v49 = icmp_imm ne v47, 0
@0068 [DynRexOp1icscc_ib#7083,%rbx] v49 = icmp_imm ne v47, 0
@0068 [RexOp2urm_noflags#4b6,%r10] v50 = bint.i32 v49
@0076 [Op1r_ib#83,%rdx] v57 = iadd_imm.i32 v23, -65
@0079 [Op1r_ib#40c1,%rdx] v59 = ishl_imm v57, 24
@007c [Op1r_ib#70c1,%rdx] v61 = sshr_imm v59, 24
@0080 [Op1r_id#4081,%rdx] v63 = band_imm v61, 255
[Op1rcmp_ib#7083,%rflags] v98 = ifcmp_imm v63, 26
@0076 [DynRexOp1r_ib#83,%rdx] v57 = iadd_imm.i32 v23, -65
@0079 [DynRexOp1r_ib#40c1,%rdx] v59 = ishl_imm v57, 24
@007c [DynRexOp1r_ib#70c1,%rdx] v61 = sshr_imm v59, 24
@0080 [DynRexOp1r_id#4081,%rdx] v63 = band_imm v61, 255
[DynRexOp1rcmp_ib#7083,%rflags] v98 = ifcmp_imm v63, 26
@0084 [RexOp1rmov#89] regmove v47, %rdi -> %rbx
@0084 [Op1brib#70] brif sge v98, ebb8
@0084 [-] fallthrough ebb11
ebb11:
[RexOp1umr#89,%rdx] v103 = copy.i32 v29
[DynRexOp1umr#89,%rdx] v103 = copy.i32 v29
@0088 [Op1jmpb#eb] jump ebb7(v29, v10, v21, v103)
ebb8:
[Op1umr#89,%rdx] v104 = copy.i32 v27
[DynRexOp1umr#89,%rdx] v104 = copy.i32 v27
@008d [RexOp1rmov#89] regmove v104, %rdx -> %r9
@008d [RexOp1rmov#89] regmove.i32 v27, %rsi -> %rdx
@008d [-] fallthrough ebb7(v104, v10, v21, v27)
ebb7(v67: i32 [%r9], v79: i32 [%rax], v81: i32 [%r8], v87: i32 [%rdx]):
@0091 [RexOp1r_id#4081,%r9] v71 = band_imm v67, 255
@0094 [RexOp1r_ib#40c1,%r9] v73 = ishl_imm v71, 24
@0097 [RexOp1r_ib#70c1,%r9] v75 = sshr_imm v73, 24
@0098 [RexOp1icscc#39,%rbx] v76 = icmp.i32 eq v47, v75
@0098 [Op2urm_noflags_abcd#4b6,%rbx] v77 = bint.i32 v76
@0099 [RexOp1rr#21,%r10] v78 = band.i32 v50, v77
@009a [RexOp1tjccb#74] brz v78, ebb9
@009a [-] fallthrough ebb12
@0091 [DynRexOp1r_id#4081,%r9] v71 = band_imm v67, 255
@0094 [DynRexOp1r_ib#40c1,%r9] v73 = ishl_imm v71, 24
@0097 [DynRexOp1r_ib#70c1,%r9] v75 = sshr_imm v73, 24
@0098 [DynRexOp1icscc#39,%rbx] v76 = icmp.i32 eq v47, v75
@0098 [Op2urm_noflags_abcd#4b6,%rbx] v77 = bint.i32 v76
@0099 [DynRexOp1rr#21,%r10] v78 = band.i32 v50, v77
@009a [DynRexOp1tjccb#74] brz v78, ebb9
@009a [-] fallthrough ebb12
ebb12:
[RexOp1umr#89,%rcx] v99 = copy v81
[Op1umr#89,%rdx] v100 = copy v79
[DynRexOp1umr#89,%rcx] v99 = copy v81
[DynRexOp1umr#89,%rdx] v100 = copy v79
@00a4 [RexOp1rmov#89] regmove v100, %rdx -> %rdi
@00a4 [RexOp1rmov#89] regmove v99, %rcx -> %rsi
@00a4 [Op1jmpd#e9] jump ebb3(v100, v99); bin: e9 ffffff2d
@@ -118,9 +118,9 @@ function u0:2691(i32 [%rdi], i32 [%rsi], i64 vmctx [%r14]) -> i64 uext [%rax] ba
@00a7 [-] fallthrough ebb4
ebb4:
@00ad [Op1r_id#4081,%rcx] v86 = band_imm.i32 v84, 255
@00b3 [Op1r_id#4081,%rdx] v89 = band_imm.i32 v87, 255
@00b4 [Op1rr#29,%rcx] v90 = isub v86, v89
@00ad [DynRexOp1r_id#4081,%rcx] v86 = band_imm.i32 v84, 255
@00b3 [DynRexOp1r_id#4081,%rdx] v89 = band_imm.i32 v87, 255
@00b4 [DynRexOp1rr#29,%rcx] v90 = isub v86, v89
@00b5 [-] fallthrough ebb2(v90)
ebb2(v5: i32 [%rcx]):

2
cranelift/filetests/filetests/isa/x86/shrink-multiple-uses.clif
View File

@@ -4,7 +4,7 @@ target x86_64
function %test_multiple_uses(i32 [%rdi]) -> i32 {
ebb0(v0: i32 [%rdi]):
[Op1rcmp_ib#7083,%rflags] v3 = ifcmp_imm v0, 0
[DynRexOp1rcmp_ib#7083,%rflags] v3 = ifcmp_imm v0, 0
[Op2seti_abcd#490,%rax] v1 = trueif eq v3
[RexOp2urm_noflags#4b6,%rax] v2 = bint.i32 v1
[Op1brib#70] brif eq v3, ebb1