From: Simon Pilgrim Date: Sun, 17 Oct 2021 17:51:15 +0000 (+0100) Subject: [CostModel][X86] Add div/rem by negative power-of-2 constants X-Git-Tag: upstream/15.0.7~28406 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=dbf5dc89306964bb613463cbec1637c6f720fbed;p=platform%2Fupstream%2Fllvm.git [CostModel][X86] Add div/rem by negative power-of-2 constants We have backend optimizations for these (like we do for power-of-2 divisions), but currently the costmodel doesn't match them --- diff --git a/llvm/test/Analysis/CostModel/X86/div.ll b/llvm/test/Analysis/CostModel/X86/div.ll index 9c9269b..b2df053 100644 --- a/llvm/test/Analysis/CostModel/X86/div.ll +++ b/llvm/test/Analysis/CostModel/X86/div.ll @@ -1428,3 +1428,612 @@ define i32 @udiv_uniformconstpow2() { ret i32 undef } + +define i32 @sdiv_constnegpow2() { +; SSE2-LABEL: 'sdiv_constnegpow2' +; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; SSE2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V4i32 = sdiv <4 x i32> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = sdiv <8 x i32> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = sdiv <16 x i32> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i8 = sdiv <32 x i8> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V64i8 = sdiv <64 x i8> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; SSSE3-LABEL: 'sdiv_constnegpow2' +; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; SSSE3-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; SSSE3-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V4i32 = sdiv <4 x i32> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = sdiv <8 x i32> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = sdiv <16 x i32> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i8 = sdiv <32 x i8> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V64i8 = sdiv <64 x i8> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; SSE42-LABEL: 'sdiv_constnegpow2' +; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = sdiv <8 x i32> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = sdiv <16 x i32> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i8 = sdiv <32 x i8> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V64i8 = sdiv <64 x i8> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX1-LABEL: 'sdiv_constnegpow2' +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = sdiv <8 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = sdiv <16 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = sdiv <16 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = sdiv <32 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = sdiv <32 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = sdiv <64 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'sdiv_constnegpow2' +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i32 = sdiv <16 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = sdiv <32 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = sdiv <32 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64i8 = sdiv <64 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512F-LABEL: 'sdiv_constnegpow2' +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = sdiv <16 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = sdiv <32 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = sdiv <32 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64i8 = sdiv <64 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512BW-LABEL: 'sdiv_constnegpow2' +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = sdiv <16 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = sdiv <32 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = sdiv <32 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64i8 = sdiv <64 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; SLM-LABEL: 'sdiv_constnegpow2' +; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; SLM-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = sdiv <8 x i32> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = sdiv <16 x i32> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; SLM-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i8 = sdiv <32 x i8> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V64i8 = sdiv <64 x i8> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; BTVER2-LABEL: 'sdiv_constnegpow2' +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = sdiv <8 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = sdiv <16 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = sdiv <16 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = sdiv <32 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = sdiv <32 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = sdiv <64 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; + %I64 = sdiv i64 undef, -16 + %V2i64 = sdiv <2 x i64> undef, + %V4i64 = sdiv <4 x i64> undef, + %V8i64 = sdiv <8 x i64> undef, + + %I32 = sdiv i32 undef, -16 + %V4i32 = sdiv <4 x i32> undef, + %V8i32 = sdiv <8 x i32> undef, + %V16i32 = sdiv <16 x i32> undef, + + %I16 = sdiv i16 undef, -16 + %V8i16 = sdiv <8 x i16> undef, + %V16i16 = sdiv <16 x i16> undef, + %V32i16 = sdiv <32 x i16> undef, + + %I8 = sdiv i8 undef, -16 + %V16i8 = sdiv <16 x i8> undef, + %V32i8 = sdiv <32 x i8> undef, + %V64i8 = sdiv <64 x i8> undef, + + ret i32 undef +} + +define i32 @udiv_constnegpow2() { +; SSE-LABEL: 'udiv_constnegpow2' +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = udiv <8 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = udiv <16 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = udiv <16 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = udiv <32 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i8 = udiv <32 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V64i8 = udiv <64 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX1-LABEL: 'udiv_constnegpow2' +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = udiv <8 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = udiv <16 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = udiv <16 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = udiv <32 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = udiv <32 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = udiv <64 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'udiv_constnegpow2' +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i32 = udiv <16 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = udiv <32 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = udiv <32 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64i8 = udiv <64 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512F-LABEL: 'udiv_constnegpow2' +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = udiv <16 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = udiv <32 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = udiv <32 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64i8 = udiv <64 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512BW-LABEL: 'udiv_constnegpow2' +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = udiv <16 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = udiv <32 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = udiv <32 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64i8 = udiv <64 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; BTVER2-LABEL: 'udiv_constnegpow2' +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = udiv <8 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = udiv <16 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = udiv <16 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = udiv <32 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = udiv <32 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = udiv <64 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; + %I64 = udiv i64 undef, -16 + %V2i64 = udiv <2 x i64> undef, + %V4i64 = udiv <4 x i64> undef, + %V8i64 = udiv <8 x i64> undef, + + %I32 = udiv i32 undef, -16 + %V4i32 = udiv <4 x i32> undef, + %V8i32 = udiv <8 x i32> undef, + %V16i32 = udiv <16 x i32> undef, + + %I16 = udiv i16 undef, -16 + %V8i16 = udiv <8 x i16> undef, + %V16i16 = udiv <16 x i16> undef, + %V32i16 = udiv <32 x i16> undef, + + %I8 = udiv i8 undef, -16 + %V16i8 = udiv <16 x i8> undef, + %V32i8 = udiv <32 x i8> undef, + %V64i8 = udiv <64 x i8> undef, + + ret i32 undef +} + +define i32 @sdiv_uniformconstnegpow2() { +; SSE-LABEL: 'sdiv_uniformconstnegpow2' +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = sdiv <4 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8i32 = sdiv <8 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16i32 = sdiv <16 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i8 = sdiv <32 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V64i8 = sdiv <64 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX1-LABEL: 'sdiv_uniformconstnegpow2' +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = sdiv <4 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8i32 = sdiv <8 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16i32 = sdiv <16 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = sdiv <16 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = sdiv <32 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = sdiv <32 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = sdiv <64 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'sdiv_uniformconstnegpow2' +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = sdiv <4 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i32 = sdiv <8 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i32 = sdiv <16 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = sdiv <32 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = sdiv <32 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64i8 = sdiv <64 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512F-LABEL: 'sdiv_uniformconstnegpow2' +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = sdiv <4 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i32 = sdiv <8 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = sdiv <16 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = sdiv <32 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = sdiv <32 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64i8 = sdiv <64 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512BW-LABEL: 'sdiv_uniformconstnegpow2' +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = sdiv <4 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i32 = sdiv <8 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = sdiv <16 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = sdiv <32 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = sdiv <32 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64i8 = sdiv <64 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; BTVER2-LABEL: 'sdiv_uniformconstnegpow2' +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = sdiv <4 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8i32 = sdiv <8 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16i32 = sdiv <16 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = sdiv <16 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = sdiv <32 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = sdiv <16 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = sdiv <32 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = sdiv <64 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; + %I64 = sdiv i64 undef, -16 + %V2i64 = sdiv <2 x i64> undef, + %V4i64 = sdiv <4 x i64> undef, + %V8i64 = sdiv <8 x i64> undef, + + %I32 = sdiv i32 undef, -16 + %V4i32 = sdiv <4 x i32> undef, + %V8i32 = sdiv <8 x i32> undef, + %V16i32 = sdiv <16 x i32> undef, + + %I16 = sdiv i16 undef, -16 + %V8i16 = sdiv <8 x i16> undef, + %V16i16 = sdiv <16 x i16> undef, + %V32i16 = sdiv <32 x i16> undef, + + %I8 = sdiv i8 undef, -16 + %V16i8 = sdiv <16 x i8> undef, + %V32i8 = sdiv <32 x i8> undef, + %V64i8 = sdiv <64 x i8> undef, + + ret i32 undef +} + +define i32 @udiv_uniformconstnegpow2() { +; SSE-LABEL: 'udiv_uniformconstnegpow2' +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i32 = udiv <4 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i32 = udiv <8 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16i32 = udiv <16 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = udiv <16 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = udiv <32 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i8 = udiv <32 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V64i8 = udiv <64 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX1-LABEL: 'udiv_uniformconstnegpow2' +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i32 = udiv <4 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8i32 = udiv <8 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16i32 = udiv <16 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = udiv <16 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = udiv <32 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = udiv <32 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = udiv <64 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'udiv_uniformconstnegpow2' +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i32 = udiv <4 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i32 = udiv <8 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16i32 = udiv <16 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = udiv <32 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = udiv <32 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64i8 = udiv <64 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512F-LABEL: 'udiv_uniformconstnegpow2' +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i32 = udiv <4 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i32 = udiv <8 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16i32 = udiv <16 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = udiv <32 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = udiv <32 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64i8 = udiv <64 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512BW-LABEL: 'udiv_uniformconstnegpow2' +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i32 = udiv <4 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i32 = udiv <8 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16i32 = udiv <16 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = udiv <32 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i8 = udiv <32 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64i8 = udiv <64 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; BTVER2-LABEL: 'udiv_uniformconstnegpow2' +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i32 = udiv <4 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8i32 = udiv <8 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16i32 = udiv <16 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = udiv <16 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = udiv <32 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i8 = udiv <16 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = udiv <32 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = udiv <64 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; + %I64 = udiv i64 undef, -16 + %V2i64 = udiv <2 x i64> undef, + %V4i64 = udiv <4 x i64> undef, + %V8i64 = udiv <8 x i64> undef, + + %I32 = udiv i32 undef, -16 + %V4i32 = udiv <4 x i32> undef, + %V8i32 = udiv <8 x i32> undef, + %V16i32 = udiv <16 x i32> undef, + + %I16 = udiv i16 undef, -16 + %V8i16 = udiv <8 x i16> undef, + %V16i16 = udiv <16 x i16> undef, + %V32i16 = udiv <32 x i16> undef, + + %I8 = udiv i8 undef, -16 + %V16i8 = udiv <16 x i8> undef, + %V32i8 = udiv <32 x i8> undef, + %V64i8 = udiv <64 x i8> undef, + + ret i32 undef +} diff --git a/llvm/test/Analysis/CostModel/X86/rem.ll b/llvm/test/Analysis/CostModel/X86/rem.ll index 8b09e42..6abfa67 100644 --- a/llvm/test/Analysis/CostModel/X86/rem.ll +++ b/llvm/test/Analysis/CostModel/X86/rem.ll @@ -1314,3 +1314,631 @@ define i32 @urem_uniformconstpow2() { ret i32 undef } + +define i32 @srem_constnegpow2() { +; SSE2-LABEL: 'srem_constnegpow2' +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; SSE2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V4i32 = srem <4 x i32> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V8i32 = srem <8 x i32> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V16i32 = srem <16 x i32> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i16 = srem <16 x i16> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i16 = srem <32 x i16> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = srem <32 x i8> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = srem <64 x i8> undef, +; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; SSSE3-LABEL: 'srem_constnegpow2' +; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; SSSE3-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; SSSE3-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V4i32 = srem <4 x i32> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V8i32 = srem <8 x i32> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V16i32 = srem <16 x i32> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i16 = srem <16 x i16> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i16 = srem <32 x i16> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; SSSE3-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = srem <32 x i8> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = srem <64 x i8> undef, +; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; SSE42-LABEL: 'srem_constnegpow2' +; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; SSE42-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = srem <4 x i32> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V8i32 = srem <8 x i32> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V16i32 = srem <16 x i32> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i16 = srem <16 x i16> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i16 = srem <32 x i16> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = srem <32 x i8> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = srem <64 x i8> undef, +; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX1-LABEL: 'srem_constnegpow2' +; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = srem <4 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = srem <8 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = srem <16 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = srem <16 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = srem <32 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32i8 = srem <32 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V64i8 = srem <64 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'srem_constnegpow2' +; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = srem <4 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = srem <8 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16i32 = srem <16 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = srem <16 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = srem <32 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = srem <32 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64i8 = srem <64 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512F-LABEL: 'srem_constnegpow2' +; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = srem <4 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = srem <8 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16i32 = srem <16 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = srem <16 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = srem <32 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = srem <32 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64i8 = srem <64 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512BW-LABEL: 'srem_constnegpow2' +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = srem <4 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = srem <8 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16i32 = srem <16 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = srem <16 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i16 = srem <32 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = srem <32 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V64i8 = srem <64 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; SLM-LABEL: 'srem_constnegpow2' +; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; SLM-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = srem <4 x i32> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V8i32 = srem <8 x i32> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V16i32 = srem <16 x i32> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; SLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i16 = srem <16 x i16> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i16 = srem <32 x i16> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = srem <32 x i8> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = srem <64 x i8> undef, +; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; GLM-LABEL: 'srem_constnegpow2' +; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; GLM-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; GLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = srem <4 x i32> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V8i32 = srem <8 x i32> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V16i32 = srem <16 x i32> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i16 = srem <16 x i16> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i16 = srem <32 x i16> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = srem <32 x i8> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = srem <64 x i8> undef, +; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; BTVER2-LABEL: 'srem_constnegpow2' +; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = srem <4 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = srem <8 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = srem <16 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = srem <16 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = srem <32 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32i8 = srem <32 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V64i8 = srem <64 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; + %I64 = srem i64 undef, -16 + %V2i64 = srem <2 x i64> undef, + %V4i64 = srem <4 x i64> undef, + %V8i64 = srem <8 x i64> undef, + + %I32 = srem i32 undef, -16 + %V4i32 = srem <4 x i32> undef, + %V8i32 = srem <8 x i32> undef, + %V16i32 = srem <16 x i32> undef, + + %I16 = srem i16 undef, -16 + %V8i16 = srem <8 x i16> undef, + %V16i16 = srem <16 x i16> undef, + %V32i16 = srem <32 x i16> undef, + + %I8 = srem i8 undef, -16 + %V16i8 = srem <16 x i8> undef, + %V32i8 = srem <32 x i8> undef, + %V64i8 = srem <64 x i8> undef, + + ret i32 undef +} + +define i32 @urem_constnegpow2() { +; SSE-LABEL: 'urem_constnegpow2' +; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = urem <4 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V8i32 = urem <8 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V16i32 = urem <16 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i16 = urem <16 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i16 = urem <32 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = urem <32 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = urem <64 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX1-LABEL: 'urem_constnegpow2' +; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = urem <4 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V8i32 = urem <8 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V16i32 = urem <16 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = urem <16 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = urem <32 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32i8 = urem <32 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V64i8 = urem <64 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'urem_constnegpow2' +; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = urem <4 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = urem <8 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16i32 = urem <16 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = urem <16 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = urem <32 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = urem <32 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64i8 = urem <64 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512F-LABEL: 'urem_constnegpow2' +; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = urem <4 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = urem <8 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16i32 = urem <16 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = urem <16 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = urem <32 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = urem <32 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64i8 = urem <64 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512BW-LABEL: 'urem_constnegpow2' +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = urem <4 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = urem <8 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16i32 = urem <16 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = urem <16 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i16 = urem <32 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = urem <32 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V64i8 = urem <64 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; BTVER2-LABEL: 'urem_constnegpow2' +; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i32 = urem <4 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V8i32 = urem <8 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V16i32 = urem <16 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = urem <16 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = urem <32 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32i8 = urem <32 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V64i8 = urem <64 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; + %I64 = urem i64 undef, -16 + %V2i64 = urem <2 x i64> undef, + %V4i64 = urem <4 x i64> undef, + %V8i64 = urem <8 x i64> undef, + + %I32 = urem i32 undef, -16 + %V4i32 = urem <4 x i32> undef, + %V8i32 = urem <8 x i32> undef, + %V16i32 = urem <16 x i32> undef, + + %I16 = urem i16 undef, -16 + %V8i16 = urem <8 x i16> undef, + %V16i16 = urem <16 x i16> undef, + %V32i16 = urem <32 x i16> undef, + + %I8 = urem i8 undef, -16 + %V16i8 = urem <16 x i8> undef, + %V32i8 = urem <32 x i8> undef, + %V64i8 = urem <64 x i8> undef, + + ret i32 undef +} + +define i32 @srem_uniformconstnegpow2() { +; SSE-LABEL: 'srem_uniformconstnegpow2' +; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4i32 = srem <4 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8i32 = srem <8 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16i32 = srem <16 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i16 = srem <16 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i16 = srem <32 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = srem <32 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = srem <64 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX1-LABEL: 'srem_uniformconstnegpow2' +; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4i32 = srem <4 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8i32 = srem <8 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V16i32 = srem <16 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = srem <16 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = srem <32 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32i8 = srem <32 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V64i8 = srem <64 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'srem_uniformconstnegpow2' +; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4i32 = srem <4 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i32 = srem <8 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i32 = srem <16 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = srem <16 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = srem <32 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = srem <32 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64i8 = srem <64 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512F-LABEL: 'srem_uniformconstnegpow2' +; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4i32 = srem <4 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i32 = srem <8 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i32 = srem <16 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = srem <16 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = srem <32 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = srem <32 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64i8 = srem <64 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512BW-LABEL: 'srem_uniformconstnegpow2' +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4i32 = srem <4 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i32 = srem <8 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i32 = srem <16 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = srem <16 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i16 = srem <32 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = srem <32 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V64i8 = srem <64 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; BTVER2-LABEL: 'srem_uniformconstnegpow2' +; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = srem i64 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = srem <2 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = srem <4 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = srem <8 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = srem i32 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4i32 = srem <4 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8i32 = srem <8 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V16i32 = srem <16 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = srem i16 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = srem <8 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = srem <16 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = srem <32 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = srem i8 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = srem <16 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32i8 = srem <32 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V64i8 = srem <64 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; + %I64 = srem i64 undef, -16 + %V2i64 = srem <2 x i64> undef, + %V4i64 = srem <4 x i64> undef, + %V8i64 = srem <8 x i64> undef, + + %I32 = srem i32 undef, -16 + %V4i32 = srem <4 x i32> undef, + %V8i32 = srem <8 x i32> undef, + %V16i32 = srem <16 x i32> undef, + + %I16 = srem i16 undef, -16 + %V8i16 = srem <8 x i16> undef, + %V16i16 = srem <16 x i16> undef, + %V32i16 = srem <32 x i16> undef, + + %I8 = srem i8 undef, -16 + %V16i8 = srem <16 x i8> undef, + %V32i8 = srem <32 x i8> undef, + %V64i8 = srem <64 x i8> undef, + + ret i32 undef +} + +define i32 @urem_uniformconstnegpow2() { +; SSE-LABEL: 'urem_uniformconstnegpow2' +; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = urem <4 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8i32 = urem <8 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16i32 = urem <16 x i32> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i16 = urem <16 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i16 = urem <32 x i16> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = urem <32 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = urem <64 x i8> undef, +; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX1-LABEL: 'urem_uniformconstnegpow2' +; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = urem <4 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8i32 = urem <8 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16i32 = urem <16 x i32> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = urem <16 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = urem <32 x i16> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32i8 = urem <32 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V64i8 = urem <64 x i8> undef, +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'urem_uniformconstnegpow2' +; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = urem <4 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i32 = urem <8 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i32 = urem <16 x i32> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = urem <16 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = urem <32 x i16> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = urem <32 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64i8 = urem <64 x i8> undef, +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512F-LABEL: 'urem_uniformconstnegpow2' +; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = urem <4 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i32 = urem <8 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i32 = urem <16 x i32> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = urem <16 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = urem <32 x i16> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = urem <32 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64i8 = urem <64 x i8> undef, +; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX512BW-LABEL: 'urem_uniformconstnegpow2' +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = urem <4 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i32 = urem <8 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i32 = urem <16 x i32> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = urem <16 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i16 = urem <32 x i16> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i8 = urem <32 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V64i8 = urem <64 x i8> undef, +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; BTVER2-LABEL: 'urem_uniformconstnegpow2' +; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = urem i64 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V2i64 = urem <2 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V4i64 = urem <4 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V8i64 = urem <8 x i64> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = urem i32 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = urem <4 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8i32 = urem <8 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16i32 = urem <16 x i32> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = urem i16 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i16 = urem <8 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = urem <16 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = urem <32 x i16> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = urem i8 undef, -16 +; BTVER2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = urem <16 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32i8 = urem <32 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V64i8 = urem <64 x i8> undef, +; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; + %I64 = urem i64 undef, -16 + %V2i64 = urem <2 x i64> undef, + %V4i64 = urem <4 x i64> undef, + %V8i64 = urem <8 x i64> undef, + + %I32 = urem i32 undef, -16 + %V4i32 = urem <4 x i32> undef, + %V8i32 = urem <8 x i32> undef, + %V16i32 = urem <16 x i32> undef, + + %I16 = urem i16 undef, -16 + %V8i16 = urem <8 x i16> undef, + %V16i16 = urem <16 x i16> undef, + %V32i16 = urem <32 x i16> undef, + + %I8 = urem i8 undef, -16 + %V16i8 = urem <16 x i8> undef, + %V32i8 = urem <32 x i8> undef, + %V64i8 = urem <64 x i8> undef, + + ret i32 undef +}