[require] shader model < 4.0 [pixel shader] // Only the used size is reserved. // Note that these matrices are column-major. float4x2 a; float4x2 b; float4 main() : sv_target { float4x2 c = a; float4x2 d = c + b; return float4(d[0].x, d[1].x, d[2].x, d[3].x); } // Registers: // // Name Reg Size // ------------ ----- ---- // a c0 1 // b c1 1 // [test] uniform 0 float4 1 2 3 4 uniform 4 float4 100 200 300 400 draw quad probe (0, 0) rgba (101, 202, 303, 404) [pixel shader] // Size 2 always are allocated before size 1 float4x2 a; float4x2 b; float4 main() : sv_target { return float4(a[0].x, a[1].x, a[2].x, b[0].y); } // Registers: // // Name Reg Size // ------------ ----- ---- // b c0 2 // a c2 1 // [test] uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 draw quad probe (0, 0) rgba (21, 22, 23, 11) [pixel shader] // Size 3 always allocated before size 2, probably means that they are allocated in decreasing used size. float4x3 a; float4x3 b; float4 main() : sv_target { return float4(a[0].y, b[0].z, 0, 0); } // Registers: // // Name Reg Size // ------------ ----- ---- // b c0 3 // a c3 2 // [test] uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 uniform 12 float4 31 32 33 34 uniform 16 float4 41 42 43 44 draw quad probe (0, 0) rgba (41, 21, 0, 0) [pixel shader] // Test same used size, how a comes before b. float4x3 a; float4x3 b; float4 main() : sv_target { return float4(a[0].z, b[0].z, 0, 0); } // Registers: // // Name Reg Size // ------------ ----- ---- // a c0 3 // b c3 3 // [test] uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 uniform 12 float4 31 32 33 34 uniform 16 float4 41 42 43 44 uniform 20 float4 51 52 53 54 draw quad probe (0, 0) rgba (21, 51, 0, 0) [pixel shader] // Declaration order is more important than instruction order to break ties. float4x3 zzz; float4x3 aaa; float4 main() : sv_target { return float4(aaa[0].z, zzz[0].z, 0, 0); } // Registers: // // Name Reg Size // ------------ ----- ---- // zzz c0 3 // aaa c3 3 // [test] uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 uniform 12 float4 31 32 33 34 uniform 16 float4 41 42 43 44 uniform 20 float4 51 52 53 54 draw quad probe (0, 0) rgba (51, 21, 0, 0) [pixel shader] // Arrays follow the same rules float2 zzz[2], aaa[4]; float4 main() : sv_target { return float4(aaa[1], zzz[1]); } // Registers: // // Name Reg Size // ------------ ----- ---- // zzz c0 2 // aaa c2 2 // [test] uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 uniform 12 float4 31 32 33 34 draw quad probe (0, 0) rgba (31, 32, 11, 12) [pixel shader] struct apple { float2 a; float3x2 b; float4 c[2]; }; struct apple zzz, hhh, aaa; float4 main() : sv_target { return float4(zzz.c[0].x, hhh.b[1].x, aaa.a); } // Registers: // // Name Reg Size // ------------ ----- ---- // zzz c0 4 // hhh c4 2 // aaa c6 1 // [test] uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 uniform 12 float4 31 32 33 34 uniform 16 float4 41 42 43 44 uniform 20 float4 51 52 53 54 uniform 24 float4 61 62 63 64 draw quad probe (0, 0) rgba (31, 52, 61, 62) [pixel shader] // Constant buffer variables are treated indistinctively. float4x4 size3; float4x4 unused; cbuffer buff { float4x4 size1; float4x4 size4; }; float4 main() : sv_target { return float4(size1[3].x, size4[0].w, size3[2].z, 0); } // Registers: // // Name Reg Size // ------------ ----- ---- // size4 c0 4 // size3 c4 3 // size1 c7 1 [test] uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 uniform 12 float4 31 32 33 34 uniform 16 float4 41 42 43 44 uniform 20 float4 51 52 53 54 uniform 24 float4 61 62 63 64 uniform 28 float4 71 72 73 74 draw quad probe (0, 0) rgba (74, 31, 63, 0) [pixel shader] typedef float2x2 matrix_t; typedef row_major matrix_t row_matrix_t; typedef column_major matrix_t col_matrix_t; uniform row_matrix_t m1; uniform col_matrix_t m2; float4 main() : COLOR { float4 ret; ret.xy = m1[0]; ret.zw = m2[0]; return ret; } // Registers: // // Name Reg Size // ------------ ----- ---- // m2 c0 2 // m1 c2 1 // [test] uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 draw quad probe (0, 0) rgba (21, 22, 1, 11) [pixel shader] // If there is a register reservation, the whole variable size is reserved. float a[4] : register(c1); float b[4]; float4 main() : sv_target { return float4(a[0], b[1], 0, 0); } // Registers: // // Name Reg Size // ------------ ----- ---- // a c1 1 // b c5 2 // [test] uniform 0 float 0 uniform 4 float 1 uniform 8 float 2 uniform 12 float 3 uniform 16 float 4 uniform 20 float 5 uniform 24 float 6 draw quad probe (0, 0) rgba (1, 6, 0, 0) [pixel shader] // If there is a register reservation, the whole variable size is reserved. float4x4 a : register(c1); float b[4]; float4 main() : sv_target { return float4(a[0].x, b[1], 0, 0); } // Registers: // // Name Reg Size // ------------ ----- ---- // a c1 1 // b c5 2 // [test] uniform 0 float 0 uniform 4 float 1 uniform 8 float 2 uniform 12 float 3 uniform 16 float 4 uniform 20 float 5 uniform 24 float 6 draw quad probe (0, 0) rgba (1, 6, 0, 0) [pixel shader] // If there is a register reservation, the whole variable size is reserved, even if unused. float4x4 unused : register(c1); float b[4]; float4 main() : sv_target { return float4(0, b[1], 0, 0); } // Registers: // // Name Reg Size // ------------ ----- ---- // b c5 2 // [test] uniform 0 float 0 uniform 4 float 1 uniform 8 float 2 uniform 12 float 3 uniform 16 float 4 uniform 20 float 5 uniform 24 float 6 draw quad probe (0, 0) rgba (0, 6, 0, 0) [pixel shader] float unused[2] : register(c2); // will create a gap in c0. float size3[3]; // cannot be placed in c0. float size2[2]; // can be placed in c0. float4 main() : sv_target { float4 res; res.x = size3[2]; res.y = size2[1]; res.z = 0; res.w = 0; return res; } // Registers: // // Name Reg Size // ------------ ----- ---- // size2 c0 2 // size3 c4 3 [test] uniform 0 float 0 uniform 4 float 1 uniform 8 float 2 uniform 12 float 3 uniform 16 float 4 uniform 20 float 5 uniform 24 float 6 draw quad probe (0, 0) rgba (6, 1, 0, 0) [pixel shader todo] // Relative addressing extends the allocation size only up to the array's size. float idx; struct { float dyn[5]; float unused; } a; float4 main() : sv_target { return a.dyn[idx]; } // Registers: // // Name Reg Size // ------------ ----- ---- // a c0 5 // idx c5 1 // [test] uniform 0 float 0 uniform 4 float 1 uniform 8 float 2 uniform 12 float 3 uniform 16 float 4 uniform 20 float 3 todo draw quad probe (0, 0) rgba (3, 3, 3, 3) uniform 20 float 1 todo draw quad probe (0, 0) rgba (1, 1, 1, 1)