freedreno/a4xx: formats updates/fixes

Update formats table with new formats that Ilia has figured out, and fix
sampling from srgb texture and integer vbo's.

Signed-off-by: Rob Clark <robclark@freedesktop.org>

3 files changed, 84 insertions, 32 deletions

diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
index 847a6753d36..7a244d8a4c5 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
@@ -328,6 +328,7 @@ fd4_emit_vertex_bufs(struct fd_ringbuffer *ring, struct fd4_emit *emit)
 			bool switchnext = (i != last) ||
 					(vertex_regid != regid(63, 0)) ||
 					(instance_regid != regid(63, 0));
+			bool isint = util_format_is_pure_integer(pfmt);
 			uint32_t fs = util_format_get_blocksize(pfmt);
 			uint32_t off = vb->buffer_offset + elem->src_offset;
 			uint32_t size = fd_bo_size(rsc->bo) - off;
@@ -350,6 +351,7 @@ fd4_emit_vertex_bufs(struct fd_ringbuffer *ring, struct fd4_emit *emit)
 					A4XX_VFD_DECODE_INSTR_REGID(vp->inputs[i].regid) |
 					A4XX_VFD_DECODE_INSTR_SHIFTCNT(fs) |
 					A4XX_VFD_DECODE_INSTR_LASTCOMPVALID |
+					COND(isint, A4XX_VFD_DECODE_INSTR_INT) |
 					COND(switchnext, A4XX_VFD_DECODE_INSTR_SWITCHNEXT));
 
 			total_in += vp->inputs[i].ncomp;
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_format.c b/src/gallium/drivers/freedreno/a4xx/fd4_format.c
index 9cff1340ee7..29abe0b0cc3 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_format.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_format.c
@@ -78,7 +78,7 @@ struct fd4_format {
 
 static struct fd4_format formats[PIPE_FORMAT_COUNT] = {
 	/* 8-bit */
-	VT(R8_UNORM,   8_UNORM, NONE,     WZYX),
+	VT(R8_UNORM,   8_UNORM, R8_UNORM, WZYX),
 	V_(R8_SNORM,   8_SNORM, NONE,     WZYX),
 	V_(R8_UINT,    8_UINT,  NONE,     WZYX),
 	V_(R8_SINT,    8_SINT,  NONE,     WZYX),
@@ -86,29 +86,38 @@ static struct fd4_format formats[PIPE_FORMAT_COUNT] = {
 	V_(R8_SSCALED, 8_UINT,  NONE,     WZYX),
 
 	_T(A8_UNORM,   8_UNORM, A8_UNORM, WZYX),
-	_T(L8_UNORM,   8_UNORM, NONE,     WZYX),
+	_T(L8_UNORM,   8_UNORM, R8_UNORM, WZYX),
 	_T(I8_UNORM,   8_UNORM, NONE,     WZYX),
 
 	/* 16-bit */
 	V_(R16_UNORM,   16_UNORM, NONE,     WZYX),
 	V_(R16_SNORM,   16_SNORM, NONE,     WZYX),
-	V_(R16_UINT,    16_UINT,  NONE,     WZYX),
-	V_(R16_SINT,    16_SINT,  NONE,     WZYX),
+	VT(R16_UINT,    16_UINT,  R16_UINT, WZYX),
+	VT(R16_SINT,    16_SINT,  R16_SINT, WZYX),
 	V_(R16_USCALED, 16_UINT,  NONE,     WZYX),
 	V_(R16_SSCALED, 16_UINT,  NONE,     WZYX),
 	VT(R16_FLOAT,   16_FLOAT, NONE,     WZYX),
 
-	VT(R8G8_UNORM,   8_8_UNORM, NONE,       WZYX),
-	V_(R8G8_SNORM,   8_8_SNORM, NONE,       WZYX),
-	V_(R8G8_UINT,    8_8_UINT,  NONE,       WZYX),
-	V_(R8G8_SINT,    8_8_SINT,  NONE,       WZYX),
+	_T(A16_UINT,    16_UINT,  NONE,     WZYX),
+	_T(A16_SINT,    16_SINT,  NONE,     WZYX),
+	_T(L16_UINT,    16_UINT,  NONE,     WZYX),
+	_T(L16_SINT,    16_SINT,  NONE,     WZYX),
+	_T(I16_UINT,    16_UINT,  NONE,     WZYX),
+	_T(I16_SINT,    16_SINT,  NONE,     WZYX),
+
+	VT(R8G8_UNORM,   8_8_UNORM, R8G8_UNORM, WZYX),
+	VT(R8G8_SNORM,   8_8_SNORM, R8G8_SNORM, WZYX),
+	VT(R8G8_UINT,    8_8_UINT,  NONE,       WZYX),
+	VT(R8G8_SINT,    8_8_SINT,  NONE,       WZYX),
 	V_(R8G8_USCALED, 8_8_UINT,  NONE,       WZYX),
 	V_(R8G8_SSCALED, 8_8_SINT,  NONE,       WZYX),
 
-	/*_T(B5G6R5_UNORM,   5_6_5_UNORM,   R5G6B5_UNORM,   WXYZ),*/
-	_T(B5G5R5A1_UNORM, 5_5_5_1_UNORM, NONE, WXYZ),
-	_T(B5G5R5X1_UNORM, 5_5_5_1_UNORM, NONE, WXYZ),
-	_T(B4G4R4A4_UNORM, 4_4_4_4_UNORM, NONE, WXYZ),
+	_T(L8A8_UINT,    8_8_UINT,  NONE,       WZYX),
+	_T(L8A8_SINT,    8_8_SINT,  NONE,       WZYX),
+
+	_T(B5G5R5A1_UNORM, 5_5_5_1_UNORM, R5G5B5A1_UNORM, WXYZ),
+	_T(B5G5R5X1_UNORM, 5_5_5_1_UNORM, R5G5B5A1_UNORM, WXYZ),
+	_T(B4G4R4A4_UNORM, 4_4_4_4_UNORM, R4G4B4A4_UNORM, WXYZ),
 
 	/* 24-bit */
 	V_(R8G8B8_UNORM,   8_8_8_UNORM, NONE, WZYX),
@@ -119,48 +128,67 @@ static struct fd4_format formats[PIPE_FORMAT_COUNT] = {
 	V_(R8G8B8_SSCALED, 8_8_8_SINT,  NONE, WZYX),
 
 	/* 32-bit */
+	VT(R32_UINT,    32_UINT,  R32_UINT, WZYX),
+	VT(R32_SINT,    32_SINT,  R32_SINT, WZYX),
+	V_(R32_USCALED, 32_UINT,  NONE,     WZYX),
+	V_(R32_SSCALED, 32_UINT,  NONE,     WZYX),
 	VT(R32_FLOAT,   32_FLOAT, NONE,     WZYX),
 	V_(R32_FIXED,   32_FIXED, NONE,     WZYX),
 
+	_T(A32_UINT,    32_UINT,  NONE,     WZYX),
+	_T(A32_SINT,    32_SINT,  NONE,     WZYX),
+	_T(L32_UINT,    32_UINT,  NONE,     WZYX),
+	_T(L32_SINT,    32_SINT,  NONE,     WZYX),
+	_T(I32_UINT,    32_UINT,  NONE,     WZYX),
+	_T(I32_SINT,    32_SINT,  NONE,     WZYX),
+
 	V_(R16G16_UNORM,   16_16_UNORM, NONE,        WZYX),
 	V_(R16G16_SNORM,   16_16_SNORM, NONE,        WZYX),
-	V_(R16G16_UINT,    16_16_UINT,  NONE,        WZYX),
-	V_(R16G16_SINT,    16_16_SINT,  NONE,        WZYX),
+	VT(R16G16_UINT,    16_16_UINT,  R16G16_UINT, WZYX),
+	VT(R16G16_SINT,    16_16_SINT,  R16G16_SINT, WZYX),
 	V_(R16G16_USCALED, 16_16_UINT,  NONE,        WZYX),
 	V_(R16G16_SSCALED, 16_16_SINT,  NONE,        WZYX),
 	VT(R16G16_FLOAT,   16_16_FLOAT, NONE,        WZYX),
 
+	_T(L16A16_UINT,    16_16_UINT,  NONE,        WZYX),
+	_T(L16A16_SINT,    16_16_SINT,  NONE,        WZYX),
+
 	VT(R8G8B8A8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 	_T(R8G8B8X8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
-	_T(R8G8B8A8_SRGB,    8_8_8_8_UNORM, NONE,           WZYX),
-	_T(R8G8B8X8_SRGB,    8_8_8_8_UNORM, NONE,           WZYX),
-	V_(R8G8B8A8_SNORM,   8_8_8_8_SNORM, NONE,           WZYX),
-	V_(R8G8B8A8_UINT,    8_8_8_8_UINT,  NONE,           WZYX),
-	V_(R8G8B8A8_SINT,    8_8_8_8_SINT,  NONE,           WZYX),
+	_T(R8G8B8A8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
+	_T(R8G8B8X8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
+	VT(R8G8B8A8_SNORM,   8_8_8_8_SNORM, R8G8B8A8_SNORM, WZYX),
+	VT(R8G8B8A8_UINT,    8_8_8_8_UINT,  R8G8B8A8_UINT,  WZYX),
+	VT(R8G8B8A8_SINT,    8_8_8_8_SINT,  R8G8B8A8_SINT,  WZYX),
 	V_(R8G8B8A8_USCALED, 8_8_8_8_UINT,  NONE,           WZYX),
 	V_(R8G8B8A8_SSCALED, 8_8_8_8_SINT,  NONE,           WZYX),
 
 	VT(B8G8R8A8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 	_T(B8G8R8X8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
-	VT(B8G8R8A8_SRGB,    8_8_8_8_UNORM, NONE,           WXYZ),
-	_T(B8G8R8X8_SRGB,    8_8_8_8_UNORM, NONE,           WXYZ),
+	VT(B8G8R8A8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
+	_T(B8G8R8X8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 
 	VT(A8B8G8R8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 	_T(X8B8G8R8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
-	_T(A8B8G8R8_SRGB,    8_8_8_8_UNORM, NONE,           XYZW),
-	_T(X8B8G8R8_SRGB,    8_8_8_8_UNORM, NONE,           XYZW),
+	_T(A8B8G8R8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
+	_T(X8B8G8R8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 
 	VT(A8R8G8B8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 	_T(X8R8G8B8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
-	_T(A8R8G8B8_SRGB,    8_8_8_8_UNORM, NONE,           ZYXW),
-	_T(X8R8G8B8_SRGB,    8_8_8_8_UNORM, NONE,           ZYXW),
+	_T(A8R8G8B8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
+	_T(X8R8G8B8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 
-	V_(R10G10B10A2_UNORM,   10_10_10_2_UNORM, NONE/*R10G10B10A2_UNORM*/, WZYX),
+	VT(R10G10B10A2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WZYX),
+	VT(B10G10R10A2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WXYZ),
+	_T(B10G10R10X2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WXYZ),
 	V_(R10G10B10A2_SNORM,   10_10_10_2_SNORM, NONE,              WZYX),
 	V_(R10G10B10A2_UINT,    10_10_10_2_UINT,  NONE,              WZYX),
 	V_(R10G10B10A2_USCALED, 10_10_10_2_UINT,  NONE,              WZYX),
 	V_(R10G10B10A2_SSCALED, 10_10_10_2_SINT,  NONE,              WZYX),
 
+	_T(R11G11B10_FLOAT, 11_11_10_FLOAT, R11G11B10_FLOAT, WZYX),
+	_T(R9G9B9E5_FLOAT,  9_9_9_E5_FLOAT, NONE,            WZYX),
+
 	_T(Z24X8_UNORM,       X8Z24_UNORM, R8G8B8A8_UNORM, WZYX),
 	_T(Z24_UNORM_S8_UINT, X8Z24_UNORM, R8G8B8A8_UNORM, WZYX),
 	/*_T(Z32_FLOAT,         Z32_FLOAT,   R8G8B8A8_UNORM, WZYX),*/
@@ -177,23 +205,42 @@ static struct fd4_format formats[PIPE_FORMAT_COUNT] = {
 	/* 64-bit */
 	V_(R16G16B16A16_UNORM,   16_16_16_16_UNORM, NONE,               WZYX),
 	V_(R16G16B16A16_SNORM,   16_16_16_16_SNORM, NONE,               WZYX),
-	V_(R16G16B16A16_UINT,    16_16_16_16_UINT,  NONE,               WZYX),
-	V_(R16G16B16A16_SINT,    16_16_16_16_SINT,  NONE,               WZYX),
+	VT(R16G16B16A16_UINT,    16_16_16_16_UINT,  R16G16B16A16_UINT,  WZYX),
+	_T(R16G16B16X16_UINT,    16_16_16_16_UINT,  R16G16B16A16_UINT,  WZYX),
+	VT(R16G16B16A16_SINT,    16_16_16_16_SINT,  R16G16B16A16_SINT,  WZYX),
+	_T(R16G16B16X16_SINT,    16_16_16_16_SINT,  R16G16B16A16_SINT,  WZYX),
 	V_(R16G16B16A16_USCALED, 16_16_16_16_UINT,  NONE,               WZYX),
 	V_(R16G16B16A16_SSCALED, 16_16_16_16_SINT,  NONE,               WZYX),
-	VT(R16G16B16A16_FLOAT,   16_16_16_16_FLOAT, NONE,               WZYX),
-	_T(R16G16B16X16_FLOAT,   16_16_16_16_FLOAT, NONE,               WZYX),
+	VT(R16G16B16A16_FLOAT,   16_16_16_16_FLOAT, R16G16B16A16_FLOAT, WZYX),
+	_T(R16G16B16X16_FLOAT,   16_16_16_16_FLOAT, R16G16B16A16_FLOAT, WZYX),
 
+	VT(R32G32_UINT,    32_32_UINT,  R32G32_UINT, WZYX),
+	VT(R32G32_SINT,    32_32_SINT,  R32G32_SINT, WZYX),
+	V_(R32G32_USCALED, 32_32_UINT,  NONE,        WZYX),
+	V_(R32G32_SSCALED, 32_32_SINT,  NONE,        WZYX),
 	VT(R32G32_FLOAT,   32_32_FLOAT, NONE,        WZYX),
 	V_(R32G32_FIXED,   32_32_FIXED, NONE,        WZYX),
 
+	_T(L32A32_UINT,    32_32_UINT,  NONE,        WZYX),
+	_T(L32A32_SINT,    32_32_SINT,  NONE,        WZYX),
+
 	/* 96-bit */
+	V_(R32G32B32_UINT,    32_32_32_UINT,  NONE, WZYX),
+	V_(R32G32B32_SINT,    32_32_32_SINT,  NONE, WZYX),
+	V_(R32G32B32_USCALED, 32_32_32_UINT,  NONE, WZYX),
+	V_(R32G32B32_SSCALED, 32_32_32_SINT,  NONE, WZYX),
 	V_(R32G32B32_FLOAT,   32_32_32_FLOAT, NONE, WZYX),
 	V_(R32G32B32_FIXED,   32_32_32_FIXED, NONE, WZYX),
 
 	/* 128-bit */
-	VT(R32G32B32A32_FLOAT,   32_32_32_32_FLOAT, NONE,               WZYX),
-	_T(R32G32B32X32_FLOAT,   32_32_32_32_FLOAT, NONE,               WZYX),
+	VT(R32G32B32A32_UINT,    32_32_32_32_UINT,  R32G32B32A32_UINT,  WZYX),
+	_T(R32G32B32X32_UINT,    32_32_32_32_UINT,  R32G32B32A32_UINT,  WZYX),
+	VT(R32G32B32A32_SINT,    32_32_32_32_SINT,  R32G32B32A32_SINT,  WZYX),
+	_T(R32G32B32X32_SINT,    32_32_32_32_SINT,  R32G32B32A32_SINT,  WZYX),
+	V_(R32G32B32A32_USCALED, 32_32_32_32_UINT,  NONE,               WZYX),
+	V_(R32G32B32A32_SSCALED, 32_32_32_32_SINT,  NONE,               WZYX),
+	VT(R32G32B32A32_FLOAT,   32_32_32_32_FLOAT, R32G32B32A32_FLOAT, WZYX),
+	_T(R32G32B32X32_FLOAT,   32_32_32_32_FLOAT, R32G32B32A32_FLOAT, WZYX),
 	V_(R32G32B32A32_FIXED,   32_32_32_32_FIXED, NONE,               WZYX),
 };
 
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_texture.c b/src/gallium/drivers/freedreno/a4xx/fd4_texture.c
index f2cc33b1a63..6ba25d0816d 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_texture.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_texture.c
@@ -169,6 +169,9 @@ fd4_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 		fd4_tex_swiz(cso->format, cso->swizzle_r, cso->swizzle_g,
 				cso->swizzle_b, cso->swizzle_a);
 
+	if (util_format_is_srgb(cso->format))
+		so->texconst0 |= A4XX_TEX_CONST_0_SRGB;
+
 	so->texconst1 =
 		A4XX_TEX_CONST_1_WIDTH(u_minify(prsc->width0, lvl)) |
 		A4XX_TEX_CONST_1_HEIGHT(u_minify(prsc->height0, lvl));