X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fauxiliary%2Fgallivm%2Flp_bld_format_soa.c;h=0a57b3ce794c6a9680829f1208ac0144afafaab8;hb=4a72d859b4f8d0444eb7f38606d59d7ddc9ea8fa;hp=9f405921b0a51f355b48fad74114e14c295ce22b;hpb=7071eefdb2ef2a1f644a2bbed9685847b60ff6c4;p=mesa.git diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c index 9f405921b0a..0a57b3ce794 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c @@ -37,6 +37,7 @@ #include "lp_bld_conv.h" #include "lp_bld_swizzle.h" #include "lp_bld_gather.h" +#include "lp_bld_debug.h" #include "lp_bld_format.h" @@ -96,12 +97,13 @@ lp_build_format_swizzle_soa(const struct util_format_description *format_desc, * \param rgba_out returns the SoA R,G,B,A vectors */ void -lp_build_unpack_rgba_soa(LLVMBuilderRef builder, +lp_build_unpack_rgba_soa(struct gallivm_state *gallivm, const struct util_format_description *format_desc, struct lp_type type, LLVMValueRef packed, LLVMValueRef rgba_out[4]) { + LLVMBuilderRef builder = gallivm->builder; struct lp_build_context bld; LLVMValueRef inputs[4]; unsigned start; @@ -115,7 +117,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, assert(type.floating); assert(type.width == 32); - lp_build_context_init(&bld, builder, type); + lp_build_context_init(&bld, gallivm, type); /* Decode the input vector components */ start = 0; @@ -128,7 +130,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, switch(format_desc->channel[chan].type) { case UTIL_FORMAT_TYPE_VOID: - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); break; case UTIL_FORMAT_TYPE_UNSIGNED: @@ -137,7 +139,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, */ if (start) { - input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(type, start), ""); + input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(gallivm, type, start), ""); } /* @@ -146,7 +148,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, if (stop < format_desc->block.bits) { unsigned mask = ((unsigned long long)1 << width) - 1; - input = LLVMBuildAnd(builder, input, lp_build_const_int_vec(type, mask), ""); + input = LLVMBuildAnd(builder, input, lp_build_const_int_vec(gallivm, type, mask), ""); } /* @@ -155,14 +157,15 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, if (type.floating) { if(format_desc->channel[chan].normalized) - input = lp_build_unsigned_norm_to_float(builder, width, type, input); + input = lp_build_unsigned_norm_to_float(gallivm, width, type, input); else - input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), ""); + input = LLVMBuildSIToFP(builder, input, + lp_build_vec_type(gallivm, type), ""); } else { /* FIXME */ assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); } break; @@ -174,7 +177,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, if (stop < type.width) { unsigned bits = type.width - stop; - LLVMValueRef bits_val = lp_build_const_int_vec(type, bits); + LLVMValueRef bits_val = lp_build_const_int_vec(gallivm, type, bits); input = LLVMBuildShl(builder, input, bits_val, ""); } @@ -184,7 +187,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, if (format_desc->channel[chan].size < type.width) { unsigned bits = type.width - format_desc->channel[chan].size; - LLVMValueRef bits_val = lp_build_const_int_vec(type, bits); + LLVMValueRef bits_val = lp_build_const_int_vec(gallivm, type, bits); input = LLVMBuildAShr(builder, input, bits_val, ""); } @@ -193,17 +196,17 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, */ if (type.floating) { - input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), ""); + input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(gallivm, type), ""); if (format_desc->channel[chan].normalized) { double scale = 1.0 / ((1 << (format_desc->channel[chan].size - 1)) - 1); - LLVMValueRef scale_val = lp_build_const_vec(type, scale); - input = LLVMBuildMul(builder, input, scale_val, ""); + LLVMValueRef scale_val = lp_build_const_vec(gallivm, type, scale); + input = LLVMBuildFMul(builder, input, scale_val, ""); } } else { /* FIXME */ assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); } break; @@ -213,32 +216,32 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, assert(start == 0); assert(stop == 32); assert(type.width == 32); - input = LLVMBuildBitCast(builder, input, lp_build_vec_type(type), ""); + input = LLVMBuildBitCast(builder, input, lp_build_vec_type(gallivm, type), ""); } else { /* FIXME */ assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); } break; case UTIL_FORMAT_TYPE_FIXED: if (type.floating) { double scale = 1.0 / ((1 << (format_desc->channel[chan].size/2)) - 1); - LLVMValueRef scale_val = lp_build_const_vec(type, scale); - input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(type), ""); - input = LLVMBuildMul(builder, input, scale_val, ""); + LLVMValueRef scale_val = lp_build_const_vec(gallivm, type, scale); + input = LLVMBuildSIToFP(builder, input, lp_build_vec_type(gallivm, type), ""); + input = LLVMBuildFMul(builder, input, scale_val, ""); } else { /* FIXME */ assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); } break; default: assert(0); - input = lp_build_undef(type); + input = lp_build_undef(gallivm, type); break; } @@ -252,16 +255,17 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, void -lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, +lp_build_rgba8_to_f32_soa(struct gallivm_state *gallivm, struct lp_type dst_type, LLVMValueRef packed, LLVMValueRef *rgba) { - LLVMValueRef mask = lp_build_const_int_vec(dst_type, 0xff); + LLVMBuilderRef builder = gallivm->builder; + LLVMValueRef mask = lp_build_const_int_vec(gallivm, dst_type, 0xff); unsigned chan; packed = LLVMBuildBitCast(builder, packed, - lp_build_int_vec_type(dst_type), ""); + lp_build_int_vec_type(gallivm, dst_type), ""); /* Decode the input vector components */ for (chan = 0; chan < 4; ++chan) { @@ -273,12 +277,12 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, if (start) input = LLVMBuildLShr(builder, input, - lp_build_const_int_vec(dst_type, start), ""); + lp_build_const_int_vec(gallivm, dst_type, start), ""); if (stop < 32) input = LLVMBuildAnd(builder, input, mask, ""); - input = lp_build_unsigned_norm_to_float(builder, 8, dst_type, input); + input = lp_build_unsigned_norm_to_float(gallivm, 8, dst_type, input); rgba[chan] = input; } @@ -302,7 +306,7 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, * be in [0, block_width-1] and j will be in [0, block_height-1]. */ void -lp_build_fetch_rgba_soa(LLVMBuilderRef builder, +lp_build_fetch_rgba_soa(struct gallivm_state *gallivm, const struct util_format_description *format_desc, struct lp_type type, LLVMValueRef base_ptr, @@ -311,6 +315,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, LLVMValueRef j, LLVMValueRef rgba_out[4]) { + LLVMBuilderRef builder = gallivm->builder; if (format_desc->layout == UTIL_FORMAT_LAYOUT_PLAIN && (format_desc->colorspace == UTIL_FORMAT_COLORSPACE_RGB || @@ -333,7 +338,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, * gather the texels from the texture * Ex: packed = {BGRA, BGRA, BGRA, BGRA}. */ - packed = lp_build_gather(builder, + packed = lp_build_gather(gallivm, type.length, format_desc->block.bits, type.width, @@ -342,7 +347,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, /* * convert texels to float rgba */ - lp_build_unpack_rgba_soa(builder, + lp_build_unpack_rgba_soa(gallivm, format_desc, type, packed, rgba_out); @@ -363,10 +368,10 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, tmp_type.length = type.length * 4; tmp_type.norm = TRUE; - tmp = lp_build_fetch_rgba_aos(builder, format_desc, tmp_type, + tmp = lp_build_fetch_rgba_aos(gallivm, format_desc, tmp_type, base_ptr, offset, i, j); - lp_build_rgba8_to_f32_soa(builder, + lp_build_rgba8_to_f32_soa(gallivm, type, tmp, rgba_out); @@ -387,27 +392,33 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, unsigned k, chan; struct lp_type tmp_type; + if (gallivm_debug & GALLIVM_DEBUG_PERF) { + debug_printf("%s: scalar unpacking of %s\n", + __FUNCTION__, format_desc->short_name); + } + tmp_type = type; tmp_type.length = 4; for (chan = 0; chan < 4; ++chan) { - rgba_out[chan] = lp_build_undef(type); + rgba_out[chan] = lp_build_undef(gallivm, type); } /* loop over number of pixels */ for(k = 0; k < type.length; ++k) { - LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), k, 0); + LLVMValueRef index = lp_build_const_int32(gallivm, k); LLVMValueRef offset_elem; LLVMValueRef i_elem, j_elem; LLVMValueRef tmp; - offset_elem = LLVMBuildExtractElement(builder, offset, index, ""); + offset_elem = LLVMBuildExtractElement(builder, offset, + index, ""); i_elem = LLVMBuildExtractElement(builder, i, index, ""); j_elem = LLVMBuildExtractElement(builder, j, index, ""); /* Get a single float[4]={R,G,B,A} pixel */ - tmp = lp_build_fetch_rgba_aos(builder, format_desc, tmp_type, + tmp = lp_build_fetch_rgba_aos(gallivm, format_desc, tmp_type, base_ptr, offset_elem, i_elem, j_elem); @@ -416,7 +427,7 @@ lp_build_fetch_rgba_soa(LLVMBuilderRef builder, * position = 'index'. */ for (chan = 0; chan < 4; ++chan) { - LLVMValueRef chan_val = LLVMConstInt(LLVMInt32Type(), chan, 0), + LLVMValueRef chan_val = lp_build_const_int32(gallivm, chan), tmp_chan = LLVMBuildExtractElement(builder, tmp, chan_val, ""); rgba_out[chan] = LLVMBuildInsertElement(builder, rgba_out[chan], tmp_chan, index, "");