| // |
| // Copyright 2016 Google Inc. |
| // |
| // Use of this source code is governed by a BSD-style |
| // license that can be found in the LICENSE file. |
| // |
| |
| #ifndef HS_GLSL_MACROS_ONCE |
| #define HS_GLSL_MACROS_ONCE |
| |
| // |
| // Define the type based on key and val sizes |
| // |
| |
| #if HS_KEY_WORDS == 1 |
| #if HS_VAL_WORDS == 0 |
| #define HS_KEY_TYPE uint |
| #endif |
| #elif HS_KEY_WORDS == 2 // FIXME -- might want to use uint2 |
| #define HS_KEY_TYPE uint64_t // GL_ARB_gpu_shader_int64 |
| #endif |
| |
| // |
| // FYI, restrict shouldn't have any impact on these kernels and |
| // benchmarks appear to prove that true |
| // |
| |
| #define HS_RESTRICT restrict |
| |
| // |
| // |
| // |
| |
| #define HS_GLSL_BINDING(n) \ |
| layout( binding = n) |
| |
| #define HS_GLSL_WORKGROUP_SIZE(x,y,z) \ |
| layout( local_size_x = x, \ |
| local_size_y = y, \ |
| local_size_z = z) in |
| |
| // |
| // KERNEL PROTOS |
| // |
| |
| #define HS_BS_KERNEL_PROTO(slab_count,slab_count_ru_log2) \ |
| HS_GLSL_SUBGROUP_SIZE() \ |
| HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS*slab_count,1,1); \ |
| HS_GLSL_BINDING(0) writeonly buffer Vout { HS_KEY_TYPE vout[]; }; \ |
| HS_GLSL_BINDING(1) readonly buffer Vin { HS_KEY_TYPE vin[]; }; \ |
| void main() |
| |
| #define HS_BC_KERNEL_PROTO(slab_count,slab_count_log2) \ |
| HS_GLSL_SUBGROUP_SIZE() \ |
| HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS*slab_count,1,1); \ |
| HS_GLSL_BINDING(0) buffer Vout { HS_KEY_TYPE vout[]; }; \ |
| void main() |
| |
| #define HS_FM_KERNEL_PROTO(s,r) \ |
| HS_GLSL_SUBGROUP_SIZE() \ |
| HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \ |
| HS_GLSL_BINDING(0) buffer Vout { HS_KEY_TYPE vout[]; }; \ |
| void main() |
| |
| #define HS_HM_KERNEL_PROTO(s) \ |
| HS_GLSL_SUBGROUP_SIZE() \ |
| HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \ |
| HS_GLSL_BINDING(0) buffer Vout { HS_KEY_TYPE vout[]; }; \ |
| void main() |
| |
| #define HS_TRANSPOSE_KERNEL_PROTO() \ |
| HS_GLSL_SUBGROUP_SIZE() \ |
| HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \ |
| HS_GLSL_BINDING(0) buffer Vout { HS_KEY_TYPE vout[]; }; \ |
| void main() |
| |
| // |
| // BLOCK LOCAL MEMORY DECLARATION |
| // |
| |
| #define HS_BLOCK_LOCAL_MEM_DECL(width,height) \ |
| shared struct { \ |
| HS_KEY_TYPE m[width * height]; \ |
| } smem |
| |
| // |
| // BLOCK BARRIER |
| // |
| |
| #define HS_BLOCK_BARRIER() \ |
| barrier() |
| |
| // |
| // SHUFFLES |
| // |
| |
| #if (HS_KEY_WORDS == 1) |
| #define HS_SHUFFLE_CAST_TO(v) v |
| #define HS_SHUFFLE_CAST_FROM(v) v |
| #elif (HS_KEY_WORDS == 2) |
| #define HS_SHUFFLE_CAST_TO(v) uint64BitsToDouble(v) |
| #define HS_SHUFFLE_CAST_FROM(v) doubleBitsToUint64(v) |
| #endif |
| |
| #define HS_SUBGROUP_SHUFFLE(v,i) HS_SHUFFLE_CAST_FROM(subgroupShuffle(HS_SHUFFLE_CAST_TO(v),i)) |
| #define HS_SUBGROUP_SHUFFLE_XOR(v,m) HS_SHUFFLE_CAST_FROM(subgroupShuffleXor(HS_SHUFFLE_CAST_TO(v),m)) |
| #define HS_SUBGROUP_SHUFFLE_UP(v,d) HS_SHUFFLE_CAST_FROM(subgroupShuffleUp(HS_SHUFFLE_CAST_TO(v),d)) |
| #define HS_SUBGROUP_SHUFFLE_DOWN(v,d) HS_SHUFFLE_CAST_FROM(subgroupShuffleDown(HS_SHUFFLE_CAST_TO(v),d)) |
| |
| // |
| // SLAB GLOBAL |
| // |
| |
| #define HS_SLAB_GLOBAL_PREAMBLE() \ |
| const uint gmem_idx = \ |
| (gl_GlobalInvocationID.x & ~(HS_SLAB_THREADS-1)) * \ |
| HS_SLAB_HEIGHT + \ |
| (gl_LocalInvocationID.x & (HS_SLAB_THREADS-1)) |
| |
| #define HS_SLAB_GLOBAL_LOAD(extent,row_idx) \ |
| extent[gmem_idx + HS_SLAB_THREADS * row_idx] |
| |
| #define HS_SLAB_GLOBAL_STORE(row_idx,reg) \ |
| vout[gmem_idx + HS_SLAB_THREADS * row_idx] = reg |
| |
| // |
| // SLAB LOCAL |
| // |
| |
| #define HS_SLAB_LOCAL_L(offset) \ |
| smem.m[smem_l_idx + (offset)] |
| |
| #define HS_SLAB_LOCAL_R(offset) \ |
| smem.m[smem_r_idx + (offset)] |
| |
| // |
| // SLAB LOCAL VERTICAL LOADS |
| // |
| |
| #define HS_BX_LOCAL_V(offset) \ |
| smem.m[gl_LocalInvocationID.x + (offset)] |
| |
| // |
| // BLOCK SORT MERGE HORIZONTAL |
| // |
| |
| #define HS_BS_MERGE_H_PREAMBLE(slab_count) \ |
| const uint smem_l_idx = \ |
| HS_SUBGROUP_ID() * (HS_SLAB_THREADS * slab_count) + \ |
| HS_SUBGROUP_LANE_ID(); \ |
| const uint smem_r_idx = \ |
| (HS_SUBGROUP_ID() ^ 1) * (HS_SLAB_THREADS * slab_count) + \ |
| (HS_SUBGROUP_LANE_ID() ^ (HS_SLAB_THREADS - 1)) |
| |
| // |
| // BLOCK CLEAN MERGE HORIZONTAL |
| // |
| |
| #define HS_BC_MERGE_H_PREAMBLE(slab_count) \ |
| const uint gmem_l_idx = \ |
| (gl_GlobalInvocationID.x & ~(HS_SLAB_THREADS * slab_count -1)) \ |
| * HS_SLAB_HEIGHT + gl_LocalInvocationID.x; \ |
| const uint smem_l_idx = \ |
| HS_SUBGROUP_ID() * (HS_SLAB_THREADS * slab_count) + \ |
| HS_SUBGROUP_LANE_ID() |
| |
| #define HS_BC_GLOBAL_LOAD_L(slab_idx) \ |
| vout[gmem_l_idx + (HS_SLAB_THREADS * slab_idx)] |
| |
| // |
| // SLAB FLIP AND HALF PREAMBLES |
| // |
| |
| #if 0 |
| |
| #define HS_SLAB_FLIP_PREAMBLE(mask) \ |
| const uint flip_lane_idx = HS_SUBGROUP_LANE_ID() ^ mask; \ |
| const bool t_lt = HS_SUBGROUP_LANE_ID() < flip_lane_idx |
| |
| #define HS_SLAB_HALF_PREAMBLE(mask) \ |
| const uint half_lane_idx = HS_SUBGROUP_LANE_ID() ^ mask; \ |
| const bool t_lt = HS_SUBGROUP_LANE_ID() < half_lane_idx |
| |
| #else |
| |
| #define HS_SLAB_FLIP_PREAMBLE(mask) \ |
| const uint flip_lane_mask = mask; \ |
| const bool t_lt = gl_LocalInvocationID.x < (gl_LocalInvocationID.x ^ mask) |
| |
| #define HS_SLAB_HALF_PREAMBLE(mask) \ |
| const uint half_lane_mask = mask; \ |
| const bool t_lt = gl_LocalInvocationID.x < (gl_LocalInvocationID.x ^ mask) |
| |
| #endif |
| |
| // |
| // Inter-lane compare exchange |
| // |
| |
| // best on 32-bit keys |
| #define HS_CMP_XCHG_V0(a,b) \ |
| { \ |
| const HS_KEY_TYPE t = min(a,b); \ |
| b = max(a,b); \ |
| a = t; \ |
| } |
| |
| // good on Intel GEN 32-bit keys |
| #define HS_CMP_XCHG_V1(a,b) \ |
| { \ |
| const HS_KEY_TYPE tmp = a; \ |
| a = (a < b) ? a : b; \ |
| b ^= a ^ tmp; \ |
| } |
| |
| // best on 64-bit keys |
| #define HS_CMP_XCHG_V2(a,b) \ |
| if (a >= b) { \ |
| const HS_KEY_TYPE t = a; \ |
| a = b; \ |
| b = t; \ |
| } |
| |
| // ok |
| #define HS_CMP_XCHG_V3(a,b) \ |
| { \ |
| const bool ge = a >= b; \ |
| const HS_KEY_TYPE t = a; \ |
| a = ge ? b : a; \ |
| b = ge ? t : b; \ |
| } |
| |
| // |
| // The flip/half comparisons rely on a "conditional min/max": |
| // |
| // - if the flag is false, return min(a,b) |
| // - otherwise, return max(a,b) |
| // |
| // What's a little surprising is that sequence (1) is faster than (2) |
| // for 32-bit keys. |
| // |
| // I suspect either a code generation problem or that the sequence |
| // maps well to the GEN instruction set. |
| // |
| // We mostly care about 64-bit keys and unsurprisingly sequence (2) is |
| // fastest for this wider type. |
| // |
| |
| #define HS_LOGICAL_XOR() != |
| |
| // this is what you would normally use |
| #define HS_COND_MIN_MAX_V0(lt,a,b) ((a <= b) HS_LOGICAL_XOR() lt) ? b : a |
| |
| // this seems to be faster for 32-bit keys on Intel GEN |
| #define HS_COND_MIN_MAX_V1(lt,a,b) (lt ? b : a) ^ ((a ^ b) & HS_LTE_TO_MASK(a,b)) |
| |
| // |
| // Conditional inter-subgroup flip/half compare exchange |
| // |
| |
| #if 0 |
| |
| #define HS_CMP_FLIP(i,a,b) \ |
| { \ |
| const HS_KEY_TYPE ta = HS_SUBGROUP_SHUFFLE(a,flip_lane_idx); \ |
| const HS_KEY_TYPE tb = HS_SUBGROUP_SHUFFLE(b,flip_lane_idx); \ |
| a = HS_COND_MIN_MAX(t_lt,a,tb); \ |
| b = HS_COND_MIN_MAX(t_lt,b,ta); \ |
| } |
| |
| #define HS_CMP_HALF(i,a) \ |
| { \ |
| const HS_KEY_TYPE ta = HS_SUBGROUP_SHUFFLE(a,half_lane_idx); \ |
| a = HS_COND_MIN_MAX(t_lt,a,ta); \ |
| } |
| |
| #else |
| |
| #define HS_CMP_FLIP(i,a,b) \ |
| { \ |
| const HS_KEY_TYPE ta = HS_SUBGROUP_SHUFFLE_XOR(a,flip_lane_mask); \ |
| const HS_KEY_TYPE tb = HS_SUBGROUP_SHUFFLE_XOR(b,flip_lane_mask); \ |
| a = HS_COND_MIN_MAX(t_lt,a,tb); \ |
| b = HS_COND_MIN_MAX(t_lt,b,ta); \ |
| } |
| |
| #define HS_CMP_HALF(i,a) \ |
| { \ |
| const HS_KEY_TYPE ta = HS_SUBGROUP_SHUFFLE_XOR(a,half_lane_mask); \ |
| a = HS_COND_MIN_MAX(t_lt,a,ta); \ |
| } |
| |
| #endif |
| |
| // |
| // The device's comparison operator might return what we actually |
| // want. For example, it appears GEN 'cmp' returns {true:-1,false:0}. |
| // |
| |
| #define HS_CMP_IS_ZERO_ONE |
| |
| #ifdef HS_CMP_IS_ZERO_ONE |
| // OpenCL requires a {true: +1, false: 0} scalar result |
| // (a < b) -> { +1, 0 } -> NEGATE -> { 0, 0xFFFFFFFF } |
| #define HS_LTE_TO_MASK(a,b) (HS_KEY_TYPE)(-(a <= b)) |
| #define HS_CMP_TO_MASK(a) (HS_KEY_TYPE)(-a) |
| #else |
| // However, OpenCL requires { -1, 0 } for vectors |
| // (a < b) -> { 0xFFFFFFFF, 0 } |
| #define HS_LTE_TO_MASK(a,b) (a <= b) // FIXME for uint64 |
| #define HS_CMP_TO_MASK(a) (a) |
| #endif |
| |
| // |
| // The "flip-merge" and "half-merge" preambles are very similar |
| // |
| // For now, we're only using the .y dimension for the span idx |
| // |
| |
| #define HS_HM_PREAMBLE(half_span) \ |
| const uint span_idx = gl_WorkGroupID.y; \ |
| const uint span_stride = gl_NumWorkGroups.x * gl_WorkGroupSize.x; \ |
| const uint span_size = span_stride * half_span * 2; \ |
| const uint span_base = span_idx * span_size; \ |
| const uint span_off = gl_GlobalInvocationID.x; \ |
| const uint span_l = span_base + span_off |
| |
| #define HS_FM_PREAMBLE(half_span) \ |
| HS_HM_PREAMBLE(half_span); \ |
| const uint span_r = span_base + span_stride * (half_span + 1) - span_off - 1 |
| |
| // |
| // |
| // |
| |
| #define HS_XM_GLOBAL_L(stride_idx) \ |
| vout[span_l + span_stride * stride_idx] |
| |
| #define HS_XM_GLOBAL_LOAD_L(stride_idx) \ |
| HS_XM_GLOBAL_L(stride_idx) |
| |
| #define HS_XM_GLOBAL_STORE_L(stride_idx,reg) \ |
| HS_XM_GLOBAL_L(stride_idx) = reg |
| |
| #define HS_FM_GLOBAL_R(stride_idx) \ |
| vout[span_r + span_stride * stride_idx] |
| |
| #define HS_FM_GLOBAL_LOAD_R(stride_idx) \ |
| HS_FM_GLOBAL_R(stride_idx) |
| |
| #define HS_FM_GLOBAL_STORE_R(stride_idx,reg) \ |
| HS_FM_GLOBAL_R(stride_idx) = reg |
| |
| // |
| // This snarl of macros is for transposing a "slab" of sorted elements |
| // into linear order. |
| // |
| // This can occur as the last step in hs_sort() or via a custom kernel |
| // that inspects the slab and then transposes and stores it to memory. |
| // |
| // The slab format can be inspected more efficiently than a linear |
| // arrangement. |
| // |
| // The prime example is detecting when adjacent keys (in sort order) |
| // have differing high order bits ("key changes"). The index of each |
| // change is recorded to an auxilary array. |
| // |
| // A post-processing step like this needs to be able to navigate the |
| // slab and eventually transpose and store the slab in linear order. |
| // |
| |
| #define HS_TRANSPOSE_REG(prefix,row) prefix##row |
| #define HS_TRANSPOSE_DECL(prefix,row) const HS_KEY_TYPE HS_TRANSPOSE_REG(prefix,row) |
| #define HS_TRANSPOSE_PRED(level) is_lo_##level |
| |
| #define HS_TRANSPOSE_TMP_REG(prefix_curr,row_ll,row_ur) \ |
| prefix_curr##row_ll##_##row_ur |
| |
| #define HS_TRANSPOSE_TMP_DECL(prefix_curr,row_ll,row_ur) \ |
| const HS_KEY_TYPE HS_TRANSPOSE_TMP_REG(prefix_curr,row_ll,row_ur) |
| |
| #define HS_TRANSPOSE_STAGE(level) \ |
| const bool HS_TRANSPOSE_PRED(level) = \ |
| (HS_SUBGROUP_LANE_ID() & (1 << (level-1))) == 0; |
| |
| #define HS_TRANSPOSE_BLEND(prefix_prev,prefix_curr,level,row_ll,row_ur) \ |
| HS_TRANSPOSE_TMP_DECL(prefix_curr,row_ll,row_ur) = \ |
| HS_SUBGROUP_SHUFFLE_XOR(HS_TRANSPOSE_PRED(level) ? \ |
| HS_TRANSPOSE_REG(prefix_prev,row_ll) : \ |
| HS_TRANSPOSE_REG(prefix_prev,row_ur), \ |
| 1<<(level-1)); \ |
| \ |
| HS_TRANSPOSE_DECL(prefix_curr,row_ll) = \ |
| HS_TRANSPOSE_PRED(level) ? \ |
| HS_TRANSPOSE_TMP_REG(prefix_curr,row_ll,row_ur) : \ |
| HS_TRANSPOSE_REG(prefix_prev,row_ll); \ |
| \ |
| HS_TRANSPOSE_DECL(prefix_curr,row_ur) = \ |
| HS_TRANSPOSE_PRED(level) ? \ |
| HS_TRANSPOSE_REG(prefix_prev,row_ur) : \ |
| HS_TRANSPOSE_TMP_REG(prefix_curr,row_ll,row_ur); |
| |
| #define HS_TRANSPOSE_REMAP(prefix,row_from,row_to) \ |
| vout[gmem_idx + ((row_to-1) << HS_SLAB_WIDTH_LOG2)] = \ |
| HS_TRANSPOSE_REG(prefix,row_from); |
| |
| // |
| // |
| // |
| |
| #endif |
| |
| // |
| // |
| // |
